diff options
Diffstat (limited to 'third_party/rust/packed_simd/src')
142 files changed, 0 insertions, 15971 deletions
diff --git a/third_party/rust/packed_simd/src/api.rs b/third_party/rust/packed_simd/src/api.rs deleted file mode 100644 index 262fc4ee69..0000000000 --- a/third_party/rust/packed_simd/src/api.rs +++ /dev/null @@ -1,309 +0,0 @@ -//! Implements the Simd<[T; N]> APIs - -#[macro_use] -mod bitmask; -pub(crate) mod cast; -#[macro_use] -mod cmp; -#[macro_use] -mod default; -#[macro_use] -mod fmt; -#[macro_use] -mod from; -#[macro_use] -mod hash; -#[macro_use] -mod math; -#[macro_use] -mod minimal; -#[macro_use] -mod ops; -#[macro_use] -mod ptr; -#[macro_use] -mod reductions; -#[macro_use] -mod select; -#[macro_use] -mod shuffle; -#[macro_use] -mod shuffle1_dyn; -#[macro_use] -mod slice; -#[macro_use] -mod swap_bytes; -#[macro_use] -mod bit_manip; - -#[cfg(feature = "into_bits")] -pub(crate) mod into_bits; - -macro_rules! impl_i { - ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident - | $ielem_ty:ident, $ibitmask_ty:ident | $test_tt:tt | $($elem_ids:ident),* - | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { - impl_minimal_iuf!([$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | $($elem_ids),* | $(#[$doc])*); - impl_ops_vector_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_scalar_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) - ); - impl_ops_scalar_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) - ); - impl_ops_vector_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_scalar_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_rotates!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_neg!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_int_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt - ); - impl_reduction_integer_arithmetic!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - ); - impl_reduction_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - ); - impl_reduction_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | (|x|{ x as $elem_ty }) | (!(0 as $elem_ty), 0) - ); - impl_fmt_debug!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_lower_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_upper_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_octal!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_binary!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_from_array!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (1, 1)); - impl_from_vectors!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* - ); - impl_default!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_hash!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_slice_from_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_slice_write_to_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_swap_bytes!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_bit_manip!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_partial_eq!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1) - ); - impl_cmp_eq!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); - impl_cmp_vertical!( - [$elem_ty; $elem_n]: $tuple_id, $mask_ty, false, (1, 0) | $test_tt - ); - impl_cmp_partial_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); - impl_bitmask!($tuple_id | $ibitmask_ty | (-1, 0) | $test_tt); - - test_select!($elem_ty, $mask_ty, $tuple_id, (1, 2) | $test_tt); - test_cmp_partial_ord_int!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - test_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - } -} - -macro_rules! impl_u { - ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident - | $ielem_ty:ident, $ibitmask_ty:ident | $test_tt:tt | $($elem_ids:ident),* - | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { - impl_minimal_iuf!([$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | $($elem_ids),* | $(#[$doc])*); - impl_ops_vector_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_scalar_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) - ); - impl_ops_scalar_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) - ); - impl_ops_vector_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_scalar_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_rotates!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_int_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt - ); - impl_reduction_integer_arithmetic!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - ); - impl_reduction_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - ); - impl_reduction_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | (|x|{ x as $elem_ty }) | (!(0 as $elem_ty), 0) - ); - impl_fmt_debug!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_lower_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_upper_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_octal!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_binary!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_from_array!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (1, 1)); - impl_from_vectors!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* - ); - impl_default!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_hash!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_slice_from_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_slice_write_to_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_swap_bytes!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_bit_manip!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_partial_eq!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (1, 0) - ); - impl_cmp_eq!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); - impl_cmp_vertical!( - [$elem_ty; $elem_n]: $tuple_id, $mask_ty, false, (1, 0) | $test_tt - ); - impl_cmp_partial_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); - impl_bitmask!($tuple_id | $ibitmask_ty | ($ielem_ty::max_value(), 0) | - $test_tt); - - test_select!($elem_ty, $mask_ty, $tuple_id, (1, 2) | $test_tt); - test_cmp_partial_ord_int!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - test_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - } -} - -macro_rules! impl_f { - ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident - | $ielem_ty:ident | $test_tt:tt | $($elem_ids:ident),* - | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { - impl_minimal_iuf!([$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | $($elem_ids),* | $(#[$doc])*); - impl_ops_vector_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_scalar_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_neg!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_ops_vector_float_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt - ); - impl_reduction_float_arithmetic!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_reduction_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - ); - impl_fmt_debug!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_from_array!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (1., 1.)); - impl_from_vectors!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* - ); - impl_default!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_partial_eq!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (1., 0.) - ); - impl_slice_from_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_slice_write_to_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - - impl_float_consts!([$elem_ty; $elem_n]: $tuple_id); - impl_float_category!([$elem_ty; $elem_n]: $tuple_id, $mask_ty); - - // floating-point math - impl_math_float_abs!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_cos!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_exp!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_ln!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_mul_add!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_mul_adde!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_powf!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_recpre!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_rsqrte!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_sin!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_sqrt!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_sqrte!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_math_float_tanh!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_vertical!( - [$elem_ty; $elem_n]: $tuple_id, $mask_ty, false, (1., 0.) - | $test_tt - ); - - test_select!($elem_ty, $mask_ty, $tuple_id, (1., 2.) | $test_tt); - test_reduction_float_min_max!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt - ); - test_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - } -} - -macro_rules! impl_m { - ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident - | $ielem_ty:ident, $ibitmask_ty:ident - | $test_tt:tt | $($elem_ids:ident),* | From: $($from_vec_ty:ident),* - | $(#[$doc:meta])*) => { - impl_minimal_mask!( - [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | $($elem_ids),* | $(#[$doc])* - ); - impl_ops_vector_mask_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) - ); - impl_ops_scalar_mask_bitwise!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) - ); - impl_reduction_bitwise!( - [bool; $elem_n]: $tuple_id | $ielem_ty | $test_tt - | (|x|{ x != 0 }) | (true, false) - ); - impl_reduction_mask!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_fmt_debug!([bool; $elem_n]: $tuple_id | $test_tt); - impl_from_array!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt - | (crate::$elem_ty::new(true), true) - ); - impl_from_vectors!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* - ); - impl_default!([bool; $elem_n]: $tuple_id | $test_tt); - impl_cmp_partial_eq!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) - ); - impl_cmp_eq!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) - ); - impl_cmp_vertical!( - [$elem_ty; $elem_n]: $tuple_id, $tuple_id, true, (true, false) - | $test_tt - ); - impl_select!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_partial_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_cmp_ord!( - [$elem_ty; $elem_n]: $tuple_id | $test_tt | (false, true) - ); - impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - impl_bitmask!($tuple_id | $ibitmask_ty | (true, false) | $test_tt); - - test_cmp_partial_ord_mask!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - test_shuffle1_dyn_mask!([$elem_ty; $elem_n]: $tuple_id | $test_tt); - } -} - -macro_rules! impl_const_p { - ([$elem_ty:ty; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident, - $usize_ty:ident, $isize_ty:ident - | $test_tt:tt | $($elem_ids:ident),* - | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { - impl_minimal_p!( - [$elem_ty; $elem_n]: $tuple_id, $mask_ty, $usize_ty, $isize_ty - | ref_ | $test_tt | $($elem_ids),* - | (1 as $elem_ty, 0 as $elem_ty) | $(#[$doc])* - ); - impl_ptr_read!([$elem_ty; $elem_n]: $tuple_id, $mask_ty | $test_tt); - } -} - -macro_rules! impl_mut_p { - ([$elem_ty:ty; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident, - $usize_ty:ident, $isize_ty:ident - | $test_tt:tt | $($elem_ids:ident),* - | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { - impl_minimal_p!( - [$elem_ty; $elem_n]: $tuple_id, $mask_ty, $usize_ty, $isize_ty - | ref_mut_ | $test_tt | $($elem_ids),* - | (1 as $elem_ty, 0 as $elem_ty) | $(#[$doc])* - ); - impl_ptr_read!([$elem_ty; $elem_n]: $tuple_id, $mask_ty | $test_tt); - impl_ptr_write!([$elem_ty; $elem_n]: $tuple_id, $mask_ty | $test_tt); - } -} diff --git a/third_party/rust/packed_simd/src/api/bit_manip.rs b/third_party/rust/packed_simd/src/api/bit_manip.rs deleted file mode 100644 index c1e90bb0fb..0000000000 --- a/third_party/rust/packed_simd/src/api/bit_manip.rs +++ /dev/null @@ -1,129 +0,0 @@ -//! Bit manipulations. - -macro_rules! impl_bit_manip { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Returns the number of ones in the binary representation of - /// the lanes of `self`. - #[inline] - pub fn count_ones(self) -> Self { - super::codegen::bit_manip::BitManip::ctpop(self) - } - - /// Returns the number of zeros in the binary representation of - /// the lanes of `self`. - #[inline] - pub fn count_zeros(self) -> Self { - super::codegen::bit_manip::BitManip::ctpop(!self) - } - - /// Returns the number of leading zeros in the binary - /// representation of the lanes of `self`. - #[inline] - pub fn leading_zeros(self) -> Self { - super::codegen::bit_manip::BitManip::ctlz(self) - } - - /// Returns the number of trailing zeros in the binary - /// representation of the lanes of `self`. - #[inline] - pub fn trailing_zeros(self) -> Self { - super::codegen::bit_manip::BitManip::cttz(self) - } - } - - test_if! { - $test_tt: - paste::item! { - #[allow(overflowing_literals)] - pub mod [<$id _bit_manip>] { - #![allow(const_item_mutation)] - use super::*; - - const LANE_WIDTH: usize = mem::size_of::<$elem_ty>() * 8; - - macro_rules! test_func { - ($x:expr, $func:ident) => {{ - let mut actual = $x; - for i in 0..$id::lanes() { - actual = actual.replace( - i, - $x.extract(i).$func() as $elem_ty - ); - } - let expected = $x.$func(); - assert_eq!(actual, expected); - }}; - } - - const BYTES: [u8; 64] = [ - 0, 1, 2, 3, 4, 5, 6, 7, - 8, 9, 10, 11, 12, 13, 14, 15, - 16, 17, 18, 19, 20, 21, 22, 23, - 24, 25, 26, 27, 28, 29, 30, 31, - 32, 33, 34, 35, 36, 37, 38, 39, - 40, 41, 42, 43, 44, 45, 46, 47, - 48, 49, 50, 51, 52, 53, 54, 55, - 56, 57, 58, 59, 60, 61, 62, 63, - ]; - - fn load_bytes() -> $id { - let elems: &mut [$elem_ty] = unsafe { - slice::from_raw_parts_mut( - BYTES.as_mut_ptr() as *mut $elem_ty, - $id::lanes(), - ) - }; - $id::from_slice_unaligned(elems) - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn count_ones() { - test_func!($id::splat(0), count_ones); - test_func!($id::splat(!0), count_ones); - test_func!(load_bytes(), count_ones); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn count_zeros() { - test_func!($id::splat(0), count_zeros); - test_func!($id::splat(!0), count_zeros); - test_func!(load_bytes(), count_zeros); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn leading_zeros() { - test_func!($id::splat(0), leading_zeros); - test_func!($id::splat(1), leading_zeros); - // some implementations use `pshufb` which has unique - // behavior when the 8th bit is set. - test_func!($id::splat(0b1000_0010), leading_zeros); - test_func!($id::splat(!0), leading_zeros); - test_func!( - $id::splat(1 << (LANE_WIDTH - 1)), - leading_zeros - ); - test_func!(load_bytes(), leading_zeros); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn trailing_zeros() { - test_func!($id::splat(0), trailing_zeros); - test_func!($id::splat(1), trailing_zeros); - test_func!($id::splat(0b1000_0010), trailing_zeros); - test_func!($id::splat(!0), trailing_zeros); - test_func!( - $id::splat(1 << (LANE_WIDTH - 1)), - trailing_zeros - ); - test_func!(load_bytes(), trailing_zeros); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/bitmask.rs b/third_party/rust/packed_simd/src/api/bitmask.rs deleted file mode 100644 index 8f4868f328..0000000000 --- a/third_party/rust/packed_simd/src/api/bitmask.rs +++ /dev/null @@ -1,79 +0,0 @@ -//! Bitmask API - -macro_rules! impl_bitmask { - ($id:ident | $ibitmask_ty:ident | ($set:expr, $clear:expr) - | $test_tt:tt) => { - impl $id { - /// Creates a bitmask with the MSB of each vector lane. - /// - /// If the vector has less than 8 lanes, the bits that do not - /// correspond to any vector lanes are cleared. - #[inline] - pub fn bitmask(self) -> $ibitmask_ty { - unsafe { codegen::llvm::simd_bitmask(self.0) } - } - } - - test_if! { - $test_tt: - paste::item! { - #[cfg(not( - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/210 - target_endian = "big" - ))] - pub mod [<$id _bitmask>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn bitmask() { - // clear all lanes - let vec = $id::splat($clear as _); - let bitmask: $ibitmask_ty = 0; - assert_eq!(vec.bitmask(), bitmask); - - // set even lanes - let mut vec = $id::splat($clear as _); - for i in 0..$id::lanes() { - if i % 2 == 0 { - vec = vec.replace(i, $set as _); - } - } - // create bitmask with even lanes set: - let mut bitmask: $ibitmask_ty = 0; - for i in 0..$id::lanes() { - if i % 2 == 0 { - bitmask |= 1 << i; - } - } - assert_eq!(vec.bitmask(), bitmask); - - - // set odd lanes - let mut vec = $id::splat($clear as _); - for i in 0..$id::lanes() { - if i % 2 != 0 { - vec = vec.replace(i, $set as _); - } - } - // create bitmask with odd lanes set: - let mut bitmask: $ibitmask_ty = 0; - for i in 0..$id::lanes() { - if i % 2 != 0 { - bitmask |= 1 << i; - } - } - assert_eq!(vec.bitmask(), bitmask); - - // set all lanes - let vec = $id::splat($set as _); - let mut bitmask: $ibitmask_ty = 0; - for i in 0..$id::lanes() { - bitmask |= 1 << i; - } - assert_eq!(vec.bitmask(), bitmask); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/cast.rs b/third_party/rust/packed_simd/src/api/cast.rs deleted file mode 100644 index f1c32ca1a3..0000000000 --- a/third_party/rust/packed_simd/src/api/cast.rs +++ /dev/null @@ -1,108 +0,0 @@ -//! Implementation of `FromCast` and `IntoCast`. -#![allow(clippy::module_name_repetitions)] - -/// Numeric cast from `T` to `Self`. -/// -/// > Note: This is a temporary workaround until the conversion traits -/// specified > in [RFC2484] are implemented. -/// -/// Numeric cast between vectors with the same number of lanes, such that: -/// -/// * casting integer vectors whose lane types have the same size (e.g. `i32xN` -/// -> `u32xN`) is a **no-op**, -/// -/// * casting from a larger integer to a smaller integer (e.g. `u32xN` -> -/// `u8xN`) will **truncate**, -/// -/// * casting from a smaller integer to a larger integer (e.g. `u8xN` -> -/// `u32xN`) will: -/// * **zero-extend** if the source is unsigned, or -/// * **sign-extend** if the source is signed, -/// -/// * casting from a float to an integer will **round the float towards zero**, -/// -/// * casting from an integer to float will produce the floating point -/// representation of the integer, **rounding to nearest, ties to even**, -/// -/// * casting from an `f32` to an `f64` is perfect and lossless, -/// -/// * casting from an `f64` to an `f32` **rounds to nearest, ties to even**. -/// -/// [RFC2484]: https://github.com/rust-lang/rfcs/pull/2484 -pub trait FromCast<T>: crate::marker::Sized { - /// Numeric cast from `T` to `Self`. - fn from_cast(_: T) -> Self; -} - -/// Numeric cast from `Self` to `T`. -/// -/// > Note: This is a temporary workaround until the conversion traits -/// specified > in [RFC2484] are implemented. -/// -/// Numeric cast between vectors with the same number of lanes, such that: -/// -/// * casting integer vectors whose lane types have the same size (e.g. `i32xN` -/// -> `u32xN`) is a **no-op**, -/// -/// * casting from a larger integer to a smaller integer (e.g. `u32xN` -> -/// `u8xN`) will **truncate**, -/// -/// * casting from a smaller integer to a larger integer (e.g. `u8xN` -> -/// `u32xN`) will: -/// * **zero-extend** if the source is unsigned, or -/// * **sign-extend** if the source is signed, -/// -/// * casting from a float to an integer will **round the float towards zero**, -/// -/// * casting from an integer to float will produce the floating point -/// representation of the integer, **rounding to nearest, ties to even**, -/// -/// * casting from an `f32` to an `f64` is perfect and lossless, -/// -/// * casting from an `f64` to an `f32` **rounds to nearest, ties to even**. -/// -/// [RFC2484]: https://github.com/rust-lang/rfcs/pull/2484 -pub trait Cast<T>: crate::marker::Sized { - /// Numeric cast from `self` to `T`. - fn cast(self) -> T; -} - -/// `FromCast` implies `Cast`. -impl<T, U> Cast<U> for T -where - U: FromCast<T>, -{ - #[inline] - fn cast(self) -> U { - U::from_cast(self) - } -} - -/// `FromCast` and `Cast` are reflexive -impl<T> FromCast<T> for T { - #[inline] - fn from_cast(t: Self) -> Self { - t - } -} - -#[macro_use] -mod macros; - -mod v16; -pub use self::v16::*; - -mod v32; -pub use self::v32::*; - -mod v64; -pub use self::v64::*; - -mod v128; -pub use self::v128::*; - -mod v256; -pub use self::v256::*; - -mod v512; -pub use self::v512::*; diff --git a/third_party/rust/packed_simd/src/api/cast/macros.rs b/third_party/rust/packed_simd/src/api/cast/macros.rs deleted file mode 100644 index 3bb29f0b80..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/macros.rs +++ /dev/null @@ -1,82 +0,0 @@ -//! Macros implementing `FromCast` - -macro_rules! impl_from_cast_ { - ($id:ident[$test_tt:tt]: $from_ty:ident) => { - impl crate::api::cast::FromCast<$from_ty> for $id { - #[inline] - fn from_cast(x: $from_ty) -> Self { - use crate::llvm::simd_cast; - debug_assert_eq!($from_ty::lanes(), $id::lanes()); - Simd(unsafe { simd_cast(x.0) }) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _from_cast_ $from_ty>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn test() { - assert_eq!($id::lanes(), $from_ty::lanes()); - } - } - } - } - }; -} - -macro_rules! impl_from_cast { - ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { - $( - impl_from_cast_!($id[$test_tt]: $from_ty); - )* - } -} - -macro_rules! impl_from_cast_mask_ { - ($id:ident[$test_tt:tt]: $from_ty:ident) => { - impl crate::api::cast::FromCast<$from_ty> for $id { - #[inline] - fn from_cast(x: $from_ty) -> Self { - debug_assert_eq!($from_ty::lanes(), $id::lanes()); - x.ne($from_ty::default()) - .select($id::splat(true), $id::splat(false)) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _from_cast_ $from_ty>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn test() { - assert_eq!($id::lanes(), $from_ty::lanes()); - - let x = $from_ty::default(); - let m: $id = x.cast(); - assert!(m.none()); - } - } - } - } - }; -} - -macro_rules! impl_from_cast_mask { - ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { - $( - impl_from_cast_mask_!($id[$test_tt]: $from_ty); - )* - } -} - -#[allow(unused)] -macro_rules! impl_into_cast { - ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { - $( - impl_from_cast_!($from_ty[$test_tt]: $id); - )* - } -} diff --git a/third_party/rust/packed_simd/src/api/cast/v128.rs b/third_party/rust/packed_simd/src/api/cast/v128.rs deleted file mode 100644 index 2e10b97b77..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/v128.rs +++ /dev/null @@ -1,302 +0,0 @@ -//! `FromCast` and `IntoCast` implementations for portable 128-bit wide vectors -#[rustfmt::skip] - -use crate::*; - -impl_from_cast!(i8x16[test_v128]: u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); -impl_from_cast!(u8x16[test_v128]: i8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); -impl_from_cast_mask!(m8x16[test_v128]: i8x16, u8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); - -impl_from_cast!( - i16x8[test_v128]: i8x8, - u8x8, - m8x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast!( - u16x8[test_v128]: i8x8, - u8x8, - m8x8, - i16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast_mask!( - m16x8[test_v128]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); - -impl_from_cast!( - i32x4[test_v128]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - u32x4[test_v128]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - f32x4[test_v128]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast_mask!( - m32x4[test_v128]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); - -impl_from_cast!( - i64x2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - u64x2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - f64x2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast_mask!( - m64x2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); - -impl_from_cast!( - isizex2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - usizex2, - msizex2 -); -impl_from_cast!( - usizex2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - msizex2 -); -impl_from_cast_mask!( - msizex2[test_v128]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2 -); - -// FIXME[test_v128]: 64-bit single element vectors into_cast impls -impl_from_cast!(i128x1[test_v128]: u128x1, m128x1); -impl_from_cast!(u128x1[test_v128]: i128x1, m128x1); -impl_from_cast!(m128x1[test_v128]: i128x1, u128x1); diff --git a/third_party/rust/packed_simd/src/api/cast/v16.rs b/third_party/rust/packed_simd/src/api/cast/v16.rs deleted file mode 100644 index 896febacbb..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/v16.rs +++ /dev/null @@ -1,68 +0,0 @@ -//! `FromCast` and `IntoCast` implementations for portable 16-bit wide vectors -#[rustfmt::skip] - -use crate::*; - -impl_from_cast!( - i8x2[test_v16]: u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - u8x2[test_v16]: i8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast_mask!( - m8x2[test_v16]: i8x2, - u8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); diff --git a/third_party/rust/packed_simd/src/api/cast/v256.rs b/third_party/rust/packed_simd/src/api/cast/v256.rs deleted file mode 100644 index fe0c835e3c..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/v256.rs +++ /dev/null @@ -1,298 +0,0 @@ -//! `FromCast` and `IntoCast` implementations for portable 256-bit wide vectors -#[rustfmt::skip] - -use crate::*; - -impl_from_cast!(i8x32[test_v256]: u8x32, m8x32, i16x32, u16x32, m16x32); -impl_from_cast!(u8x32[test_v256]: i8x32, m8x32, i16x32, u16x32, m16x32); -impl_from_cast_mask!(m8x32[test_v256]: i8x32, u8x32, i16x32, u16x32, m16x32); - -impl_from_cast!(i16x16[test_v256]: i8x16, u8x16, m8x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); -impl_from_cast!(u16x16[test_v256]: i8x16, u8x16, m8x16, i16x16, m16x16, i32x16, u32x16, f32x16, m32x16); -impl_from_cast_mask!(m16x16[test_v256]: i8x16, u8x16, m8x16, i16x16, u16x16, i32x16, u32x16, f32x16, m32x16); - -impl_from_cast!( - i32x8[test_v256]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast!( - u32x8[test_v256]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast!( - f32x8[test_v256]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast_mask!( - m32x8[test_v256]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); - -impl_from_cast!( - i64x4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - u64x4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - f64x4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast_mask!( - m64x4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); - -impl_from_cast!( - i128x2[test_v256]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - u128x2[test_v256]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast_mask!( - m128x2[test_v256]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - m64x2, - f64x2, - i128x2, - u128x2, - isizex2, - usizex2, - msizex2 -); - -impl_from_cast!( - isizex4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - usizex4, - msizex4 -); -impl_from_cast!( - usizex4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - msizex4 -); -impl_from_cast_mask!( - msizex4[test_v256]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4 -); diff --git a/third_party/rust/packed_simd/src/api/cast/v32.rs b/third_party/rust/packed_simd/src/api/cast/v32.rs deleted file mode 100644 index 4ad1cbf74d..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/v32.rs +++ /dev/null @@ -1,132 +0,0 @@ -//! `FromCast` and `IntoCast` implementations for portable 32-bit wide vectors -#[rustfmt::skip] - -use crate::*; - -impl_from_cast!( - i8x4[test_v32]: u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - u8x4[test_v32]: i8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast_mask!( - m8x4[test_v32]: i8x4, - u8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); - -impl_from_cast!( - i16x2[test_v32]: i8x2, - u8x2, - m8x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - u16x2[test_v32]: i8x2, - u8x2, - m8x2, - i16x2, - m16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast_mask!( - m16x2[test_v32]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - i32x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); diff --git a/third_party/rust/packed_simd/src/api/cast/v512.rs b/third_party/rust/packed_simd/src/api/cast/v512.rs deleted file mode 100644 index b64605045e..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/v512.rs +++ /dev/null @@ -1,209 +0,0 @@ -//! `FromCast` and `IntoCast` implementations for portable 512-bit wide vectors -#[rustfmt::skip] - -use crate::*; - -impl_from_cast!(i8x64[test_v512]: u8x64, m8x64); -impl_from_cast!(u8x64[test_v512]: i8x64, m8x64); -impl_from_cast_mask!(m8x64[test_v512]: i8x64, u8x64); - -impl_from_cast!(i16x32[test_v512]: i8x32, u8x32, m8x32, u16x32, m16x32); -impl_from_cast!(u16x32[test_v512]: i8x32, u8x32, m8x32, i16x32, m16x32); -impl_from_cast_mask!(m16x32[test_v512]: i8x32, u8x32, m8x32, i16x32, u16x32); - -impl_from_cast!(i32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, u32x16, f32x16, m32x16); -impl_from_cast!(u32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, f32x16, m32x16); -impl_from_cast!(f32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, m32x16); -impl_from_cast_mask!(m32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16); - -impl_from_cast!( - i64x8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast!( - u64x8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast!( - f64x8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast_mask!( - m64x8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - isizex8, - usizex8, - msizex8 -); - -impl_from_cast!( - i128x4[test_v512]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - u128x4[test_v512]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast_mask!( - m128x4[test_v512]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - m64x4, - f64x4, - i128x4, - u128x4, - isizex4, - usizex4, - msizex4 -); - -impl_from_cast!( - isizex8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - usizex8, - msizex8 -); -impl_from_cast!( - usizex8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - msizex8 -); -impl_from_cast_mask!( - msizex8[test_v512]: i8x8, - u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8 -); diff --git a/third_party/rust/packed_simd/src/api/cast/v64.rs b/third_party/rust/packed_simd/src/api/cast/v64.rs deleted file mode 100644 index b23d1a4917..0000000000 --- a/third_party/rust/packed_simd/src/api/cast/v64.rs +++ /dev/null @@ -1,208 +0,0 @@ -//! `FromCast` and `IntoCast` implementations for portable 64-bit wide vectors -#[rustfmt::skip] - -use crate::*; - -impl_from_cast!( - i8x8[test_v64]: u8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast!( - u8x8[test_v64]: i8x8, - m8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); -impl_from_cast_mask!( - m8x8[test_v64]: i8x8, - u8x8, - i16x8, - u16x8, - m16x8, - i32x8, - u32x8, - f32x8, - m32x8, - i64x8, - u64x8, - f64x8, - m64x8, - isizex8, - usizex8, - msizex8 -); - -impl_from_cast!( - i16x4[test_v64]: i8x4, - u8x4, - m8x4, - u16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast!( - u16x4[test_v64]: i8x4, - u8x4, - m8x4, - i16x4, - m16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); -impl_from_cast_mask!( - m16x4[test_v64]: i8x4, - u8x4, - m8x4, - i16x4, - u16x4, - i32x4, - u32x4, - f32x4, - m32x4, - i64x4, - u64x4, - f64x4, - m64x4, - i128x4, - u128x4, - m128x4, - isizex4, - usizex4, - msizex4 -); - -impl_from_cast!( - i32x2[test_v64]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - u32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - u32x2[test_v64]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - f32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast!( - f32x2[test_v64]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - m32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); -impl_from_cast_mask!( - m32x2[test_v64]: i8x2, - u8x2, - m8x2, - i16x2, - u16x2, - m16x2, - i32x2, - u32x2, - f32x2, - i64x2, - u64x2, - f64x2, - m64x2, - i128x2, - u128x2, - m128x2, - isizex2, - usizex2, - msizex2 -); diff --git a/third_party/rust/packed_simd/src/api/cmp.rs b/third_party/rust/packed_simd/src/api/cmp.rs deleted file mode 100644 index 6d5301dddd..0000000000 --- a/third_party/rust/packed_simd/src/api/cmp.rs +++ /dev/null @@ -1,16 +0,0 @@ -//! Implement cmp traits for vector types - -#[macro_use] -mod partial_eq; - -#[macro_use] -mod eq; - -#[macro_use] -mod partial_ord; - -#[macro_use] -mod ord; - -#[macro_use] -mod vertical; diff --git a/third_party/rust/packed_simd/src/api/cmp/eq.rs b/third_party/rust/packed_simd/src/api/cmp/eq.rs deleted file mode 100644 index 3c55d0dce5..0000000000 --- a/third_party/rust/packed_simd/src/api/cmp/eq.rs +++ /dev/null @@ -1,27 +0,0 @@ -//! Implements `Eq` for vector types. - -macro_rules! impl_cmp_eq { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - impl crate::cmp::Eq for $id {} - impl crate::cmp::Eq for LexicographicallyOrdered<$id> {} - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_eq>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn eq() { - fn foo<E: crate::cmp::Eq>(_: E) {} - let a = $id::splat($false); - foo(a); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/cmp/ord.rs b/third_party/rust/packed_simd/src/api/cmp/ord.rs deleted file mode 100644 index e54ba3bfde..0000000000 --- a/third_party/rust/packed_simd/src/api/cmp/ord.rs +++ /dev/null @@ -1,43 +0,0 @@ -//! Implements `Ord` for vector types. - -macro_rules! impl_cmp_ord { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - impl $id { - /// Returns a wrapper that implements `Ord`. - #[inline] - pub fn lex_ord(&self) -> LexicographicallyOrdered<$id> { - LexicographicallyOrdered(*self) - } - } - - impl crate::cmp::Ord for LexicographicallyOrdered<$id> { - #[inline] - fn cmp(&self, other: &Self) -> crate::cmp::Ordering { - match self.partial_cmp(other) { - Some(x) => x, - None => unsafe { crate::hint::unreachable_unchecked() }, - } - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_ord>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn eq() { - fn foo<E: crate::cmp::Ord>(_: E) {} - let a = $id::splat($false); - foo(a.partial_lex_ord()); - foo(a.lex_ord()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/cmp/partial_eq.rs b/third_party/rust/packed_simd/src/api/cmp/partial_eq.rs deleted file mode 100644 index d69dd47425..0000000000 --- a/third_party/rust/packed_simd/src/api/cmp/partial_eq.rs +++ /dev/null @@ -1,65 +0,0 @@ -//! Implements `PartialEq` for vector types. - -macro_rules! impl_cmp_partial_eq { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - // FIXME: https://github.com/rust-lang-nursery/rust-clippy/issues/2892 - #[allow(clippy::partialeq_ne_impl)] - impl crate::cmp::PartialEq<$id> for $id { - #[inline] - fn eq(&self, other: &Self) -> bool { - $id::eq(*self, *other).all() - } - #[inline] - fn ne(&self, other: &Self) -> bool { - $id::ne(*self, *other).any() - } - } - - // FIXME: https://github.com/rust-lang-nursery/rust-clippy/issues/2892 - #[allow(clippy::partialeq_ne_impl)] - impl crate::cmp::PartialEq<LexicographicallyOrdered<$id>> for LexicographicallyOrdered<$id> { - #[inline] - fn eq(&self, other: &Self) -> bool { - self.0 == other.0 - } - #[inline] - fn ne(&self, other: &Self) -> bool { - self.0 != other.0 - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _cmp_PartialEq>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn partial_eq() { - let a = $id::splat($false); - let b = $id::splat($true); - - assert!(a != b); - assert!(!(a == b)); - assert!(a == a); - assert!(!(a != a)); - - if $id::lanes() > 1 { - let a = $id::splat($false).replace(0, $true); - let b = $id::splat($true); - - assert!(a != b); - assert!(!(a == b)); - assert!(a == a); - assert!(!(a != a)); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/cmp/partial_ord.rs b/third_party/rust/packed_simd/src/api/cmp/partial_ord.rs deleted file mode 100644 index 76ed9ebe4e..0000000000 --- a/third_party/rust/packed_simd/src/api/cmp/partial_ord.rs +++ /dev/null @@ -1,230 +0,0 @@ -//! Implements `PartialOrd` for vector types. -//! -//! This implements a lexicographical order. - -macro_rules! impl_cmp_partial_ord { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Returns a wrapper that implements `PartialOrd`. - #[inline] - pub fn partial_lex_ord(&self) -> LexicographicallyOrdered<$id> { - LexicographicallyOrdered(*self) - } - } - - impl crate::cmp::PartialOrd<LexicographicallyOrdered<$id>> for LexicographicallyOrdered<$id> { - #[inline] - fn partial_cmp(&self, other: &Self) -> Option<crate::cmp::Ordering> { - if PartialEq::eq(self, other) { - Some(crate::cmp::Ordering::Equal) - } else if PartialOrd::lt(self, other) { - Some(crate::cmp::Ordering::Less) - } else if PartialOrd::gt(self, other) { - Some(crate::cmp::Ordering::Greater) - } else { - None - } - } - #[inline] - fn lt(&self, other: &Self) -> bool { - let m_lt = self.0.lt(other.0); - let m_eq = self.0.eq(other.0); - for i in 0..$id::lanes() { - if m_eq.extract(i) { - continue; - } - return m_lt.extract(i); - } - false - } - #[inline] - fn le(&self, other: &Self) -> bool { - self.lt(other) | PartialEq::eq(self, other) - } - #[inline] - fn ge(&self, other: &Self) -> bool { - self.gt(other) | PartialEq::eq(self, other) - } - #[inline] - fn gt(&self, other: &Self) -> bool { - let m_gt = self.0.gt(other.0); - let m_eq = self.0.eq(other.0); - for i in 0..$id::lanes() { - if m_eq.extract(i) { - continue; - } - return m_gt.extract(i); - } - false - } - } - }; -} - -macro_rules! test_cmp_partial_ord_int { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_PartialOrd>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn partial_lex_ord() { - use crate::testing::utils::{test_cmp}; - // constant values - let a = $id::splat(0); - let b = $id::splat(1); - - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - - // variable values: a = [0, 1, 2, 3]; b = [3, 2, 1, 0] - let mut a = $id::splat(0); - let mut b = $id::splat(0); - for i in 0..$id::lanes() { - a = a.replace(i, i as $elem_ty); - b = b.replace(i, ($id::lanes() - i) as $elem_ty); - } - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - - // variable values: a = [0, 1, 2, 3]; b = [0, 1, 2, 4] - let mut b = a; - b = b.replace( - $id::lanes() - 1, - a.extract($id::lanes() - 1) + 1 as $elem_ty - ); - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - - if $id::lanes() > 2 { - // variable values a = [0, 1, 0, 0]; b = [0, 1, 2, 3] - let b = a; - let mut a = $id::splat(0); - a = a.replace(1, 1 as $elem_ty); - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - - // variable values: a = [0, 1, 2, 3]; b = [0, 1, 3, 2] - let mut b = a; - b = b.replace( - 2, a.extract($id::lanes() - 1) + 1 as $elem_ty - ); - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(crate::cmp::Ordering::Equal)); - } - } - } - } - } - }; -} - -macro_rules! test_cmp_partial_ord_mask { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_PartialOrd>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn partial_lex_ord() { - use crate::testing::utils::{test_cmp}; - use crate::cmp::Ordering; - - // constant values - let a = $id::splat(false); - let b = $id::splat(true); - - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Equal)); - - // variable values: - // a = [false, false, false, false]; - // b = [false, false, false, true] - let a = $id::splat(false); - let mut b = $id::splat(false); - b = b.replace($id::lanes() - 1, true); - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Equal)); - - // variable values: - // a = [true, true, true, false]; - // b = [true, true, true, true] - let mut a = $id::splat(true); - let b = $id::splat(true); - a = a.replace($id::lanes() - 1, false); - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Equal)); - - if $id::lanes() > 2 { - // variable values - // a = [false, true, false, false]; - // b = [false, true, true, true] - let mut a = $id::splat(false); - let mut b = $id::splat(true); - a = a.replace(1, true); - b = b.replace(0, false); - test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Less)); - test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Greater)); - test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), - Some(Ordering::Equal)); - test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), - Some(Ordering::Equal)); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/cmp/vertical.rs b/third_party/rust/packed_simd/src/api/cmp/vertical.rs deleted file mode 100644 index ea4a0d1a34..0000000000 --- a/third_party/rust/packed_simd/src/api/cmp/vertical.rs +++ /dev/null @@ -1,114 +0,0 @@ -//! Vertical (lane-wise) vector comparisons returning vector masks. - -macro_rules! impl_cmp_vertical { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident, - $mask_ty:ident, - $is_mask:expr,($true:expr, $false:expr) | $test_tt:tt - ) => { - impl $id { - /// Lane-wise equality comparison. - #[inline] - pub fn eq(self, other: Self) -> $mask_ty { - use crate::llvm::simd_eq; - Simd(unsafe { simd_eq(self.0, other.0) }) - } - - /// Lane-wise inequality comparison. - #[inline] - pub fn ne(self, other: Self) -> $mask_ty { - use crate::llvm::simd_ne; - Simd(unsafe { simd_ne(self.0, other.0) }) - } - - /// Lane-wise less-than comparison. - #[inline] - pub fn lt(self, other: Self) -> $mask_ty { - use crate::llvm::{simd_gt, simd_lt}; - if $is_mask { - Simd(unsafe { simd_gt(self.0, other.0) }) - } else { - Simd(unsafe { simd_lt(self.0, other.0) }) - } - } - - /// Lane-wise less-than-or-equals comparison. - #[inline] - pub fn le(self, other: Self) -> $mask_ty { - use crate::llvm::{simd_ge, simd_le}; - if $is_mask { - Simd(unsafe { simd_ge(self.0, other.0) }) - } else { - Simd(unsafe { simd_le(self.0, other.0) }) - } - } - - /// Lane-wise greater-than comparison. - #[inline] - pub fn gt(self, other: Self) -> $mask_ty { - use crate::llvm::{simd_gt, simd_lt}; - if $is_mask { - Simd(unsafe { simd_lt(self.0, other.0) }) - } else { - Simd(unsafe { simd_gt(self.0, other.0) }) - } - } - - /// Lane-wise greater-than-or-equals comparison. - #[inline] - pub fn ge(self, other: Self) -> $mask_ty { - use crate::llvm::{simd_ge, simd_le}; - if $is_mask { - Simd(unsafe { simd_le(self.0, other.0) }) - } else { - Simd(unsafe { simd_ge(self.0, other.0) }) - } - } - } - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_vertical>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn cmp() { - let a = $id::splat($false); - let b = $id::splat($true); - - let r = a.lt(b); - let e = $mask_ty::splat(true); - assert!(r == e); - let r = a.le(b); - assert!(r == e); - - let e = $mask_ty::splat(false); - let r = a.gt(b); - assert!(r == e); - let r = a.ge(b); - assert!(r == e); - let r = a.eq(b); - assert!(r == e); - - let mut a = a; - let mut b = b; - let mut e = e; - for i in 0..$id::lanes() { - if i % 2 == 0 { - a = a.replace(i, $false); - b = b.replace(i, $true); - e = e.replace(i, true); - } else { - a = a.replace(i, $true); - b = b.replace(i, $false); - e = e.replace(i, false); - } - } - let r = a.lt(b); - assert!(r == e); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/default.rs b/third_party/rust/packed_simd/src/api/default.rs deleted file mode 100644 index 7af55ea77a..0000000000 --- a/third_party/rust/packed_simd/src/api/default.rs +++ /dev/null @@ -1,30 +0,0 @@ -//! Implements `Default` for vector types. - -macro_rules! impl_default { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl Default for $id { - #[inline] - fn default() -> Self { - Self::splat($elem_ty::default()) - } - } - - test_if!{ - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _default>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn default() { - let a = $id::default(); - for i in 0..$id::lanes() { - assert_eq!(a.extract(i), $elem_ty::default()); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/fmt.rs b/third_party/rust/packed_simd/src/api/fmt.rs deleted file mode 100644 index f3f55c4015..0000000000 --- a/third_party/rust/packed_simd/src/api/fmt.rs +++ /dev/null @@ -1,12 +0,0 @@ -//! Implements formatting APIs - -#[macro_use] -mod debug; -#[macro_use] -mod lower_hex; -#[macro_use] -mod upper_hex; -#[macro_use] -mod octal; -#[macro_use] -mod binary; diff --git a/third_party/rust/packed_simd/src/api/fmt/binary.rs b/third_party/rust/packed_simd/src/api/fmt/binary.rs deleted file mode 100644 index 91c0825559..0000000000 --- a/third_party/rust/packed_simd/src/api/fmt/binary.rs +++ /dev/null @@ -1,54 +0,0 @@ -//! Implement Octal formatting - -macro_rules! impl_fmt_binary { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::fmt::Binary for $id { - #[allow(clippy::missing_inline_in_public_items)] - fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { - write!(f, "{}(", stringify!($id))?; - for i in 0..$elem_count { - if i > 0 { - write!(f, ", ")?; - } - self.extract(i).fmt(f)?; - } - write!(f, ")") - } - } - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _fmt_binary>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn binary() { - use arrayvec::{ArrayString,ArrayVec}; - type TinyString = ArrayString<[u8; 512]>; - - use crate::fmt::Write; - let v = $id::splat($elem_ty::default()); - let mut s = TinyString::new(); - write!(&mut s, "{:#b}", v).unwrap(); - - let mut beg = TinyString::new(); - write!(&mut beg, "{}(", stringify!($id)).unwrap(); - assert!(s.starts_with(beg.as_str())); - assert!(s.ends_with(")")); - let s: ArrayVec<[TinyString; 64]> - = s.replace(beg.as_str(), "") - .replace(")", "").split(",") - .map(|v| TinyString::from(v.trim()).unwrap()) - .collect(); - assert_eq!(s.len(), $id::lanes()); - for (index, ss) in s.into_iter().enumerate() { - let mut e = TinyString::new(); - write!(&mut e, "{:#b}", v.extract(index)).unwrap(); - assert_eq!(ss, e); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/fmt/debug.rs b/third_party/rust/packed_simd/src/api/fmt/debug.rs deleted file mode 100644 index 1e209b3bff..0000000000 --- a/third_party/rust/packed_simd/src/api/fmt/debug.rs +++ /dev/null @@ -1,60 +0,0 @@ -//! Implement debug formatting - -macro_rules! impl_fmt_debug_tests { - ([$elem_ty:ty; $elem_count:expr]: $id:ident | $test_tt:tt) => { - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _fmt_debug>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn debug() { - use arrayvec::{ArrayString,ArrayVec}; - type TinyString = ArrayString<[u8; 512]>; - - use crate::fmt::Write; - let v = $id::default(); - let mut s = TinyString::new(); - write!(&mut s, "{:?}", v).unwrap(); - - let mut beg = TinyString::new(); - write!(&mut beg, "{}(", stringify!($id)).unwrap(); - assert!(s.starts_with(beg.as_str())); - assert!(s.ends_with(")")); - let s: ArrayVec<[TinyString; 64]> - = s.replace(beg.as_str(), "") - .replace(")", "").split(",") - .map(|v| TinyString::from(v.trim()).unwrap()) - .collect(); - assert_eq!(s.len(), $id::lanes()); - for (index, ss) in s.into_iter().enumerate() { - let mut e = TinyString::new(); - write!(&mut e, "{:?}", v.extract(index)).unwrap(); - assert_eq!(ss, e); - } - } - } - } - } - }; -} - -macro_rules! impl_fmt_debug { - ([$elem_ty:ty; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::fmt::Debug for $id { - #[allow(clippy::missing_inline_in_public_items)] - fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { - write!(f, "{}(", stringify!($id))?; - for i in 0..$elem_count { - if i > 0 { - write!(f, ", ")?; - } - self.extract(i).fmt(f)?; - } - write!(f, ")") - } - } - impl_fmt_debug_tests!([$elem_ty; $elem_count]: $id | $test_tt); - }; -} diff --git a/third_party/rust/packed_simd/src/api/fmt/lower_hex.rs b/third_party/rust/packed_simd/src/api/fmt/lower_hex.rs deleted file mode 100644 index 8f11d3119b..0000000000 --- a/third_party/rust/packed_simd/src/api/fmt/lower_hex.rs +++ /dev/null @@ -1,54 +0,0 @@ -//! Implement `LowerHex` formatting - -macro_rules! impl_fmt_lower_hex { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::fmt::LowerHex for $id { - #[allow(clippy::missing_inline_in_public_items)] - fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { - write!(f, "{}(", stringify!($id))?; - for i in 0..$elem_count { - if i > 0 { - write!(f, ", ")?; - } - self.extract(i).fmt(f)?; - } - write!(f, ")") - } - } - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _fmt_lower_hex>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn lower_hex() { - use arrayvec::{ArrayString,ArrayVec}; - type TinyString = ArrayString<[u8; 512]>; - - use crate::fmt::Write; - let v = $id::splat($elem_ty::default()); - let mut s = TinyString::new(); - write!(&mut s, "{:#x}", v).unwrap(); - - let mut beg = TinyString::new(); - write!(&mut beg, "{}(", stringify!($id)).unwrap(); - assert!(s.starts_with(beg.as_str())); - assert!(s.ends_with(")")); - let s: ArrayVec<[TinyString; 64]> - = s.replace(beg.as_str(), "").replace(")", "") - .split(",") - .map(|v| TinyString::from(v.trim()).unwrap()) - .collect(); - assert_eq!(s.len(), $id::lanes()); - for (index, ss) in s.into_iter().enumerate() { - let mut e = TinyString::new(); - write!(&mut e, "{:#x}", v.extract(index)).unwrap(); - assert_eq!(ss, e); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/fmt/octal.rs b/third_party/rust/packed_simd/src/api/fmt/octal.rs deleted file mode 100644 index e708e094ce..0000000000 --- a/third_party/rust/packed_simd/src/api/fmt/octal.rs +++ /dev/null @@ -1,54 +0,0 @@ -//! Implement Octal formatting - -macro_rules! impl_fmt_octal { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::fmt::Octal for $id { - #[allow(clippy::missing_inline_in_public_items)] - fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { - write!(f, "{}(", stringify!($id))?; - for i in 0..$elem_count { - if i > 0 { - write!(f, ", ")?; - } - self.extract(i).fmt(f)?; - } - write!(f, ")") - } - } - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _fmt_octal>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn octal_hex() { - use arrayvec::{ArrayString,ArrayVec}; - type TinyString = ArrayString<[u8; 512]>; - - use crate::fmt::Write; - let v = $id::splat($elem_ty::default()); - let mut s = TinyString::new(); - write!(&mut s, "{:#o}", v).unwrap(); - - let mut beg = TinyString::new(); - write!(&mut beg, "{}(", stringify!($id)).unwrap(); - assert!(s.starts_with(beg.as_str())); - assert!(s.ends_with(")")); - let s: ArrayVec<[TinyString; 64]> - = s.replace(beg.as_str(), "").replace(")", "") - .split(",") - .map(|v| TinyString::from(v.trim()).unwrap()) - .collect(); - assert_eq!(s.len(), $id::lanes()); - for (index, ss) in s.into_iter().enumerate() { - let mut e = TinyString::new(); - write!(&mut e, "{:#o}", v.extract(index)).unwrap(); - assert_eq!(ss, e); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/fmt/upper_hex.rs b/third_party/rust/packed_simd/src/api/fmt/upper_hex.rs deleted file mode 100644 index 5ad455706b..0000000000 --- a/third_party/rust/packed_simd/src/api/fmt/upper_hex.rs +++ /dev/null @@ -1,54 +0,0 @@ -//! Implement `UpperHex` formatting - -macro_rules! impl_fmt_upper_hex { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::fmt::UpperHex for $id { - #[allow(clippy::missing_inline_in_public_items)] - fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { - write!(f, "{}(", stringify!($id))?; - for i in 0..$elem_count { - if i > 0 { - write!(f, ", ")?; - } - self.extract(i).fmt(f)?; - } - write!(f, ")") - } - } - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _fmt_upper_hex>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn upper_hex() { - use arrayvec::{ArrayString,ArrayVec}; - type TinyString = ArrayString<[u8; 512]>; - - use crate::fmt::Write; - let v = $id::splat($elem_ty::default()); - let mut s = TinyString::new(); - write!(&mut s, "{:#X}", v).unwrap(); - - let mut beg = TinyString::new(); - write!(&mut beg, "{}(", stringify!($id)).unwrap(); - assert!(s.starts_with(beg.as_str())); - assert!(s.ends_with(")")); - let s: ArrayVec<[TinyString; 64]> - = s.replace(beg.as_str(), "").replace(")", "") - .split(",") - .map(|v| TinyString::from(v.trim()).unwrap()) - .collect(); - assert_eq!(s.len(), $id::lanes()); - for (index, ss) in s.into_iter().enumerate() { - let mut e = TinyString::new(); - write!(&mut e, "{:#X}", v.extract(index)).unwrap(); - assert_eq!(ss, e); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/from.rs b/third_party/rust/packed_simd/src/api/from.rs deleted file mode 100644 index c30c4d6e21..0000000000 --- a/third_party/rust/packed_simd/src/api/from.rs +++ /dev/null @@ -1,7 +0,0 @@ -//! Implementations of the `From` and `Into` traits - -#[macro_use] -mod from_array; - -#[macro_use] -mod from_vector; diff --git a/third_party/rust/packed_simd/src/api/from/from_array.rs b/third_party/rust/packed_simd/src/api/from/from_array.rs deleted file mode 100644 index 5c7801ddaf..0000000000 --- a/third_party/rust/packed_simd/src/api/from/from_array.rs +++ /dev/null @@ -1,124 +0,0 @@ -//! Implements `From<[T; N]>` and `Into<[T; N]>` for vector types. - -macro_rules! impl_from_array { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt - | ($non_default_array:expr, $non_default_vec:expr)) => { - impl From<[$elem_ty; $elem_count]> for $id { - #[inline] - fn from(array: [$elem_ty; $elem_count]) -> Self { - union U { - array: [$elem_ty; $elem_count], - vec: $id, - } - unsafe { U { array }.vec } - } - } - - impl From<$id> for [$elem_ty; $elem_count] { - #[inline] - fn from(vec: $id) -> Self { - union U { - array: [$elem_ty; $elem_count], - vec: $id, - } - unsafe { U { vec }.array } - } - } - - // FIXME: `Into::into` is not inline, but due to - // the blanket impl in `std`, which is not - // marked `default`, we cannot override it here with - // specialization. - /* - impl Into<[$elem_ty; $elem_count]> for $id { - #[inline] - fn into(self) -> [$elem_ty; $elem_count] { - union U { - array: [$elem_ty; $elem_count], - vec: $id, - } - unsafe { U { vec: self }.array } - } - } - - impl Into<$id> for [$elem_ty; $elem_count] { - #[inline] - fn into(self) -> $id { - union U { - array: [$elem_ty; $elem_count], - vec: $id, - } - unsafe { U { array: self }.vec } - } - } - */ - - test_if! { - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - mod [<$id _from>] { - use super::*; - #[test] - #[cfg_attr(miri, ignore)] - fn array() { - let vec: $id = Default::default(); - - // FIXME: Workaround for arrays with more than 32 - // elements. - // - // Safe because we never take a reference to any - // uninitialized element. - union W { - array: [$elem_ty; $elem_count], - other: () - } - let mut array = W { other: () }; - for i in 0..$elem_count { - let default: $elem_ty = Default::default(); - // note: array.other is the active member and - // initialized so we can take a reference to it: - let p = unsafe { - &mut array.other as *mut () as *mut $elem_ty - }; - // note: default is a valid bit-pattern for - // $elem_ty: - unsafe { - crate::ptr::write(p.wrapping_add(i), default) - }; - } - // note: the array variant of the union is properly - // initialized: - let mut array = unsafe { - array.array - }; - - array[0] = $non_default_array; - let vec = vec.replace(0, $non_default_vec); - - let vec_from_array = $id::from(array); - assert_eq!(vec_from_array, vec); - let array_from_vec - = <[$elem_ty; $elem_count]>::from(vec); - // FIXME: Workaround for arrays with more than 32 - // elements. - for i in 0..$elem_count { - assert_eq!(array_from_vec[i], array[i]); - } - - let vec_from_into_array: $id = array.into(); - assert_eq!(vec_from_into_array, vec); - let array_from_into_vec: [$elem_ty; $elem_count] - = vec.into(); - // FIXME: Workaround for arrays with more than 32 - // elements. - for i in 0..$elem_count { - assert_eq!(array_from_into_vec[i], array[i]); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/from/from_vector.rs b/third_party/rust/packed_simd/src/api/from/from_vector.rs deleted file mode 100644 index 55f70016d5..0000000000 --- a/third_party/rust/packed_simd/src/api/from/from_vector.rs +++ /dev/null @@ -1,67 +0,0 @@ -//! Implements `From` and `Into` for vector types. - -macro_rules! impl_from_vector { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt - | $source:ident) => { - impl From<$source> for $id { - #[inline] - fn from(source: $source) -> Self { - fn static_assert_same_number_of_lanes<T, U>() - where - T: crate::sealed::Simd, - U: crate::sealed::Simd<LanesType = T::LanesType>, - { - } - use crate::llvm::simd_cast; - static_assert_same_number_of_lanes::<$id, $source>(); - Simd(unsafe { simd_cast(source.0) }) - } - } - - // FIXME: `Into::into` is not inline, but due to the blanket impl in - // `std`, which is not marked `default`, we cannot override it here - // with specialization. - - /* - impl Into<$id> for $source { - #[inline] - fn into(self) -> $id { - unsafe { simd_cast(self) } - } - } - */ - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _from_ $source>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from() { - assert_eq!($id::lanes(), $source::lanes()); - let source: $source = Default::default(); - let vec: $id = Default::default(); - - let e = $id::from(source); - assert_eq!(e, vec); - - let e: $id = source.into(); - assert_eq!(e, vec); - } - } - } - } - }; -} - -macro_rules! impl_from_vectors { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt - | $($source:ident),*) => { - $( - impl_from_vector!( - [$elem_ty; $elem_count]: $id | $test_tt | $source - ); - )* - } -} diff --git a/third_party/rust/packed_simd/src/api/hash.rs b/third_party/rust/packed_simd/src/api/hash.rs deleted file mode 100644 index ee80eff939..0000000000 --- a/third_party/rust/packed_simd/src/api/hash.rs +++ /dev/null @@ -1,49 +0,0 @@ -//! Implements `Hash` for vector types. - -macro_rules! impl_hash { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::hash::Hash for $id { - #[inline] - fn hash<H: crate::hash::Hasher>(&self, state: &mut H) { - unsafe { - union A { - data: [$elem_ty; $id::lanes()], - vec: $id, - } - A { vec: *self }.data.hash(state) - } - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _hash>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn hash() { - use crate::hash::{Hash, Hasher}; - #[allow(deprecated)] - use crate::hash::{SipHasher13}; - type A = [$elem_ty; $id::lanes()]; - let a: A = [42 as $elem_ty; $id::lanes()]; - assert_eq!( - crate::mem::size_of::<A>(), - crate::mem::size_of::<$id>() - ); - #[allow(deprecated)] - let mut a_hash = SipHasher13::new(); - let mut v_hash = a_hash.clone(); - a.hash(&mut a_hash); - - // Integer within mantissa^1 range. - #[allow(clippy::float_cmp)] - let v = $id::splat(42 as $elem_ty); - v.hash(&mut v_hash); - assert_eq!(a_hash.finish(), v_hash.finish()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/into_bits.rs b/third_party/rust/packed_simd/src/api/into_bits.rs deleted file mode 100644 index 03fbe4bff7..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits.rs +++ /dev/null @@ -1,59 +0,0 @@ -//! Implementation of `FromBits` and `IntoBits`. - -/// Safe lossless bitwise conversion from `T` to `Self`. -#[cfg_attr(doc_cfg, doc(cfg(feature = "into_bits")))] -pub trait FromBits<T>: crate::marker::Sized { - /// Safe lossless bitwise transmute from `T` to `Self`. - fn from_bits(t: T) -> Self; -} - -/// Safe lossless bitwise conversion from `Self` to `T`. -#[cfg_attr(doc_cfg, doc(cfg(feature = "into_bits")))] -pub trait IntoBits<T>: crate::marker::Sized { - /// Safe lossless bitwise transmute from `self` to `T`. - fn into_bits(self) -> T; -} - -/// `FromBits` implies `IntoBits`. -impl<T, U> IntoBits<U> for T -where - U: FromBits<T>, -{ - #[inline] - fn into_bits(self) -> U { - debug_assert!(crate::mem::size_of::<Self>() == crate::mem::size_of::<U>()); - U::from_bits(self) - } -} - -/// `FromBits` and `IntoBits` are reflexive -impl<T> FromBits<T> for T { - #[inline] - fn from_bits(t: Self) -> Self { - t - } -} - -#[macro_use] -mod macros; - -mod v16; -pub use self::v16::*; - -mod v32; -pub use self::v32::*; - -mod v64; -pub use self::v64::*; - -mod v128; -pub use self::v128::*; - -mod v256; -pub use self::v256::*; - -mod v512; -pub use self::v512::*; - -mod arch_specific; -pub use self::arch_specific::*; diff --git a/third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs b/third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs deleted file mode 100644 index bfac915576..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs +++ /dev/null @@ -1,345 +0,0 @@ -//! `FromBits` and `IntoBits` between portable vector types and the -//! architecture-specific vector types. -#[rustfmt::skip] - -// FIXME: MIPS FromBits/IntoBits - -#[allow(unused)] -use crate::*; - -/// This macro implements FromBits for the portable and the architecture -/// specific vector types. -/// -/// The "leaf" case is at the bottom, and the most generic case is at the top. -/// The generic case is split into smaller cases recursively. -macro_rules! impl_arch { - ([$arch_head_i:ident[$arch_head_tt:tt]: $($arch_head_ty:ident),*], - $([$arch_tail_i:ident[$arch_tail_tt:tt]: $($arch_tail_ty:ident),*]),* | - from: $($from_ty:ident),* | into: $($into_ty:ident),* | - test: $test_tt:tt) => { - impl_arch!( - [$arch_head_i[$arch_head_tt]: $($arch_head_ty),*] | - from: $($from_ty),* | - into: $($into_ty),* | - test: $test_tt - ); - impl_arch!( - $([$arch_tail_i[$arch_tail_tt]: $($arch_tail_ty),*]),* | - from: $($from_ty),* | - into: $($into_ty),* | - test: $test_tt - ); - }; - ([$arch:ident[$arch_tt:tt]: $($arch_ty:ident),*] | - from: $($from_ty:ident),* | into: $($into_ty:ident),* | - test: $test_tt:tt) => { - // note: if target is "arm", "+v7,+neon" must be enabled - // and the std library must be recompiled with them - #[cfg(any( - not(target_arch = "arm"), - all(target_feature = "v7", target_feature = "neon", - any(feature = "core_arch", libcore_neon))) - )] - // note: if target is "powerpc", "altivec" must be enabled - // and the std library must be recompiled with it - #[cfg(any( - not(target_arch = "powerpc"), - all(target_feature = "altivec", feature = "core_arch"), - ))] - #[cfg(target_arch = $arch_tt)] - use crate::arch::$arch::{ - $($arch_ty),* - }; - - #[cfg(any( - not(target_arch = "arm"), - all(target_feature = "v7", target_feature = "neon", - any(feature = "core_arch", libcore_neon))) - )] - #[cfg(any( - not(target_arch = "powerpc"), - all(target_feature = "altivec", feature = "core_arch"), - ))] - #[cfg(target_arch = $arch_tt)] - impl_arch!($($arch_ty),* | $($from_ty),* | $($into_ty),* | - test: $test_tt); - }; - ($arch_head:ident, $($arch_tail:ident),* | $($from_ty:ident),* - | $($into_ty:ident),* | test: $test_tt:tt) => { - impl_arch!($arch_head | $($from_ty),* | $($into_ty),* | - test: $test_tt); - impl_arch!($($arch_tail),* | $($from_ty),* | $($into_ty),* | - test: $test_tt); - }; - ($arch_head:ident | $($from_ty:ident),* | $($into_ty:ident),* | - test: $test_tt:tt) => { - impl_from_bits!($arch_head[$test_tt]: $($from_ty),*); - impl_into_bits!($arch_head[$test_tt]: $($into_ty),*); - }; -} - -//////////////////////////////////////////////////////////////////////////////// -// Implementations for the 64-bit wide vector types: - -// FIXME: 64-bit single element types -// FIXME: arm/aarch float16x4_t missing -impl_arch!( - [ - arm["arm"]: int8x8_t, - uint8x8_t, - poly8x8_t, - int16x4_t, - uint16x4_t, - poly16x4_t, - int32x2_t, - uint32x2_t, - float32x2_t, - int64x1_t, - uint64x1_t - ], - [ - aarch64["aarch64"]: int8x8_t, - uint8x8_t, - poly8x8_t, - int16x4_t, - uint16x4_t, - poly16x4_t, - int32x2_t, - uint32x2_t, - float32x2_t, - int64x1_t, - uint64x1_t, - float64x1_t - ] | from: i8x8, - u8x8, - m8x8, - i16x4, - u16x4, - m16x4, - i32x2, - u32x2, - f32x2, - m32x2 | into: i8x8, - u8x8, - i16x4, - u16x4, - i32x2, - u32x2, - f32x2 | test: test_v64 -); - -//////////////////////////////////////////////////////////////////////////////// -// Implementations for the 128-bit wide vector types: - -// FIXME: arm/aarch float16x8_t missing -// FIXME: ppc vector_pixel missing -// FIXME: ppc64 vector_Float16 missing -// FIXME: ppc64 vector_signed_long_long missing -// FIXME: ppc64 vector_unsigned_long_long missing -// FIXME: ppc64 vector_bool_long_long missing -// FIXME: ppc64 vector_signed___int128 missing -// FIXME: ppc64 vector_unsigned___int128 missing -impl_arch!( - [x86["x86"]: __m128, __m128i, __m128d], - [x86_64["x86_64"]: __m128, __m128i, __m128d], - [ - arm["arm"]: int8x16_t, - uint8x16_t, - poly8x16_t, - int16x8_t, - uint16x8_t, - poly16x8_t, - int32x4_t, - uint32x4_t, - float32x4_t, - int64x2_t, - uint64x2_t - ], - [ - aarch64["aarch64"]: int8x16_t, - uint8x16_t, - poly8x16_t, - int16x8_t, - uint16x8_t, - poly16x8_t, - int32x4_t, - uint32x4_t, - float32x4_t, - int64x2_t, - uint64x2_t, - float64x2_t - ], - [ - powerpc["powerpc"]: vector_signed_char, - vector_unsigned_char, - vector_signed_short, - vector_unsigned_short, - vector_signed_int, - vector_unsigned_int, - vector_float - ], - [ - powerpc64["powerpc64"]: vector_signed_char, - vector_unsigned_char, - vector_signed_short, - vector_unsigned_short, - vector_signed_int, - vector_unsigned_int, - vector_float, - vector_signed_long, - vector_unsigned_long, - vector_double - ] | from: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 | into: i8x16, - u8x16, - i16x8, - u16x8, - i32x4, - u32x4, - f32x4, - i64x2, - u64x2, - f64x2, - i128x1, - u128x1 | test: test_v128 -); - -impl_arch!( - [powerpc["powerpc"]: vector_bool_char], - [powerpc64["powerpc64"]: vector_bool_char] | from: m8x16, - m16x8, - m32x4, - m64x2, - m128x1 | into: i8x16, - u8x16, - i16x8, - u16x8, - i32x4, - u32x4, - f32x4, - i64x2, - u64x2, - f64x2, - i128x1, - u128x1, - // Masks: - m8x16 | test: test_v128 -); - -impl_arch!( - [powerpc["powerpc"]: vector_bool_short], - [powerpc64["powerpc64"]: vector_bool_short] | from: m16x8, - m32x4, - m64x2, - m128x1 | into: i8x16, - u8x16, - i16x8, - u16x8, - i32x4, - u32x4, - f32x4, - i64x2, - u64x2, - f64x2, - i128x1, - u128x1, - // Masks: - m8x16, - m16x8 | test: test_v128 -); - -impl_arch!( - [powerpc["powerpc"]: vector_bool_int], - [powerpc64["powerpc64"]: vector_bool_int] | from: m32x4, - m64x2, - m128x1 | into: i8x16, - u8x16, - i16x8, - u16x8, - i32x4, - u32x4, - f32x4, - i64x2, - u64x2, - f64x2, - i128x1, - u128x1, - // Masks: - m8x16, - m16x8, - m32x4 | test: test_v128 -); - -impl_arch!( - [powerpc64["powerpc64"]: vector_bool_long] | from: m64x2, - m128x1 | into: i8x16, - u8x16, - i16x8, - u16x8, - i32x4, - u32x4, - f32x4, - i64x2, - u64x2, - f64x2, - i128x1, - u128x1, - // Masks: - m8x16, - m16x8, - m32x4, - m64x2 | test: test_v128 -); - -//////////////////////////////////////////////////////////////////////////////// -// Implementations for the 256-bit wide vector types - -impl_arch!( - [x86["x86"]: __m256, __m256i, __m256d], - [x86_64["x86_64"]: __m256, __m256i, __m256d] | from: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 | into: i8x32, - u8x32, - i16x16, - u16x16, - i32x8, - u32x8, - f32x8, - i64x4, - u64x4, - f64x4, - i128x2, - u128x2 | test: test_v256 -); - -//////////////////////////////////////////////////////////////////////////////// -// FIXME: Implementations for the 512-bit wide vector types diff --git a/third_party/rust/packed_simd/src/api/into_bits/macros.rs b/third_party/rust/packed_simd/src/api/into_bits/macros.rs deleted file mode 100644 index 265ab34ae0..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/macros.rs +++ /dev/null @@ -1,74 +0,0 @@ -//! Macros implementing `FromBits` - -macro_rules! impl_from_bits_ { - ($id:ident[$test_tt:tt]: $from_ty:ident) => { - impl crate::api::into_bits::FromBits<$from_ty> for $id { - #[inline] - fn from_bits(x: $from_ty) -> Self { - unsafe { crate::mem::transmute(x) } - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _from_bits_ $from_ty>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn test() { - use crate::{ - ptr::{read_unaligned}, - mem::{size_of, zeroed} - }; - use crate::IntoBits; - assert_eq!(size_of::<$id>(), - size_of::<$from_ty>()); - // This is safe because we never create a reference to - // uninitialized memory: - let a: $from_ty = unsafe { zeroed() }; - - let b_0: $id = crate::FromBits::from_bits(a); - let b_1: $id = a.into_bits(); - - // Check that these are byte-wise equal, that is, - // that the bit patterns are identical: - for i in 0..size_of::<$id>() { - // This is safe because we only read initialized - // memory in bounds. Also, taking a reference to - // `b_i` is ok because the fields are initialized. - unsafe { - let b_0_v: u8 = read_unaligned( - (&b_0 as *const $id as *const u8) - .wrapping_add(i) - ); - let b_1_v: u8 = read_unaligned( - (&b_1 as *const $id as *const u8) - .wrapping_add(i) - ); - assert_eq!(b_0_v, b_1_v); - } - } - } - } - } - } - }; -} - -macro_rules! impl_from_bits { - ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { - $( - impl_from_bits_!($id[$test_tt]: $from_ty); - )* - } -} - -#[allow(unused)] -macro_rules! impl_into_bits { - ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { - $( - impl_from_bits_!($from_ty[$test_tt]: $id); - )* - } -} diff --git a/third_party/rust/packed_simd/src/api/into_bits/v128.rs b/third_party/rust/packed_simd/src/api/into_bits/v128.rs deleted file mode 100644 index 639c09c2c4..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/v128.rs +++ /dev/null @@ -1,232 +0,0 @@ -//! `FromBits` and `IntoBits` implementations for portable 128-bit wide vectors -#[rustfmt::skip] - -#[allow(unused)] // wasm_bindgen_test -use crate::*; - -impl_from_bits!( - i8x16[test_v128]: u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!( - u8x16[test_v128]: i8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!(m8x16[test_v128]: m16x8, m32x4, m64x2, m128x1); - -impl_from_bits!( - i16x8[test_v128]: i8x16, - u8x16, - m8x16, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!( - u16x8[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!(m16x8[test_v128]: m32x4, m64x2, m128x1); - -impl_from_bits!( - i32x4[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!( - u32x4[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!( - f32x4[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!(m32x4[test_v128]: m64x2, m128x1); - -impl_from_bits!( - i64x2[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - u64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!( - u64x2[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - f64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!( - f64x2[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - m64x2, - i128x1, - u128x1, - m128x1 -); -impl_from_bits!(m64x2[test_v128]: m128x1); - -impl_from_bits!( - i128x1[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - u128x1, - m128x1 -); -impl_from_bits!( - u128x1[test_v128]: i8x16, - u8x16, - m8x16, - i16x8, - u16x8, - m16x8, - i32x4, - u32x4, - f32x4, - m32x4, - i64x2, - u64x2, - f64x2, - m64x2, - i128x1, - m128x1 -); -// note: m128x1 cannot be constructed from all the other masks bit patterns in -// here diff --git a/third_party/rust/packed_simd/src/api/into_bits/v16.rs b/third_party/rust/packed_simd/src/api/into_bits/v16.rs deleted file mode 100644 index e44d0e7f9a..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/v16.rs +++ /dev/null @@ -1,9 +0,0 @@ -//! `FromBits` and `IntoBits` implementations for portable 16-bit wide vectors -#[rustfmt::skip] - -#[allow(unused)] // wasm_bindgen_test -use crate::*; - -impl_from_bits!(i8x2[test_v16]: u8x2, m8x2); -impl_from_bits!(u8x2[test_v16]: i8x2, m8x2); -// note: m8x2 cannot be constructed from all i8x2 or u8x2 bit patterns diff --git a/third_party/rust/packed_simd/src/api/into_bits/v256.rs b/third_party/rust/packed_simd/src/api/into_bits/v256.rs deleted file mode 100644 index e432bbbc9f..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/v256.rs +++ /dev/null @@ -1,232 +0,0 @@ -//! `FromBits` and `IntoBits` implementations for portable 256-bit wide vectors -#[rustfmt::skip] - -#[allow(unused)] // wasm_bindgen_test -use crate::*; - -impl_from_bits!( - i8x32[test_v256]: u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!( - u8x32[test_v256]: i8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!(m8x32[test_v256]: m16x16, m32x8, m64x4, m128x2); - -impl_from_bits!( - i16x16[test_v256]: i8x32, - u8x32, - m8x32, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!( - u16x16[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!(m16x16[test_v256]: m32x8, m64x4, m128x2); - -impl_from_bits!( - i32x8[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!( - u32x8[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!( - f32x8[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!(m32x8[test_v256]: m64x4, m128x2); - -impl_from_bits!( - i64x4[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - u64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!( - u64x4[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - f64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!( - f64x4[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - m64x4, - i128x2, - u128x2, - m128x2 -); -impl_from_bits!(m64x4[test_v256]: m128x2); - -impl_from_bits!( - i128x2[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - u128x2, - m128x2 -); -impl_from_bits!( - u128x2[test_v256]: i8x32, - u8x32, - m8x32, - i16x16, - u16x16, - m16x16, - i32x8, - u32x8, - f32x8, - m32x8, - i64x4, - u64x4, - f64x4, - m64x4, - i128x2, - m128x2 -); -// note: m128x2 cannot be constructed from all the other masks bit patterns in -// here diff --git a/third_party/rust/packed_simd/src/api/into_bits/v32.rs b/third_party/rust/packed_simd/src/api/into_bits/v32.rs deleted file mode 100644 index 5dba38a179..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/v32.rs +++ /dev/null @@ -1,13 +0,0 @@ -//! `FromBits` and `IntoBits` implementations for portable 32-bit wide vectors -#[rustfmt::skip] - -#[allow(unused)] // wasm_bindgen_test -use crate::*; - -impl_from_bits!(i8x4[test_v32]: u8x4, m8x4, i16x2, u16x2, m16x2); -impl_from_bits!(u8x4[test_v32]: i8x4, m8x4, i16x2, u16x2, m16x2); -impl_from_bits!(m8x4[test_v32]: m16x2); - -impl_from_bits!(i16x2[test_v32]: i8x4, u8x4, m8x4, u16x2, m16x2); -impl_from_bits!(u16x2[test_v32]: i8x4, u8x4, m8x4, i16x2, m16x2); -// note: m16x2 cannot be constructed from all m8x4 bit patterns diff --git a/third_party/rust/packed_simd/src/api/into_bits/v512.rs b/third_party/rust/packed_simd/src/api/into_bits/v512.rs deleted file mode 100644 index f6e9bb8bf7..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/v512.rs +++ /dev/null @@ -1,232 +0,0 @@ -//! `FromBits` and `IntoBits` implementations for portable 512-bit wide vectors -#[rustfmt::skip] - -#[allow(unused)] // wasm_bindgen_test -use crate::*; - -impl_from_bits!( - i8x64[test_v512]: u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!( - u8x64[test_v512]: i8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!(m8x64[test_v512]: m16x32, m32x16, m64x8, m128x4); - -impl_from_bits!( - i16x32[test_v512]: i8x64, - u8x64, - m8x64, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!( - u16x32[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!(m16x32[test_v512]: m32x16, m64x8, m128x4); - -impl_from_bits!( - i32x16[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!( - u32x16[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!( - f32x16[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!(m32x16[test_v512]: m64x8, m128x4); - -impl_from_bits!( - i64x8[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - u64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!( - u64x8[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - f64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!( - f64x8[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - m64x8, - i128x4, - u128x4, - m128x4 -); -impl_from_bits!(m64x8[test_v512]: m128x4); - -impl_from_bits!( - i128x4[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - u128x4, - m128x4 -); -impl_from_bits!( - u128x4[test_v512]: i8x64, - u8x64, - m8x64, - i16x32, - u16x32, - m16x32, - i32x16, - u32x16, - f32x16, - m32x16, - i64x8, - u64x8, - f64x8, - m64x8, - i128x4, - m128x4 -); -// note: m128x4 cannot be constructed from all the other masks bit patterns in -// here diff --git a/third_party/rust/packed_simd/src/api/into_bits/v64.rs b/third_party/rust/packed_simd/src/api/into_bits/v64.rs deleted file mode 100644 index 5b065f1bd5..0000000000 --- a/third_party/rust/packed_simd/src/api/into_bits/v64.rs +++ /dev/null @@ -1,18 +0,0 @@ -//! `FromBits` and `IntoBits` implementations for portable 64-bit wide vectors -#[rustfmt::skip] - -#[allow(unused)] // wasm_bindgen_test -use crate::*; - -impl_from_bits!(i8x8[test_v64]: u8x8, m8x8, i16x4, u16x4, m16x4, i32x2, u32x2, f32x2, m32x2); -impl_from_bits!(u8x8[test_v64]: i8x8, m8x8, i16x4, u16x4, m16x4, i32x2, u32x2, f32x2, m32x2); -impl_from_bits!(m8x8[test_v64]: m16x4, m32x2); - -impl_from_bits!(i16x4[test_v64]: i8x8, u8x8, m8x8, u16x4, m16x4, i32x2, u32x2, f32x2, m32x2); -impl_from_bits!(u16x4[test_v64]: i8x8, u8x8, m8x8, i16x4, m16x4, i32x2, u32x2, f32x2, m32x2); -impl_from_bits!(m16x4[test_v64]: m32x2); - -impl_from_bits!(i32x2[test_v64]: i8x8, u8x8, m8x8, i16x4, u16x4, m16x4, u32x2, f32x2, m32x2); -impl_from_bits!(u32x2[test_v64]: i8x8, u8x8, m8x8, i16x4, u16x4, m16x4, i32x2, f32x2, m32x2); -impl_from_bits!(f32x2[test_v64]: i8x8, u8x8, m8x8, i16x4, u16x4, m16x4, i32x2, u32x2, m32x2); -// note: m32x2 cannot be constructed from all m16x4 or m8x8 bit patterns diff --git a/third_party/rust/packed_simd/src/api/math.rs b/third_party/rust/packed_simd/src/api/math.rs deleted file mode 100644 index e7a8d256ba..0000000000 --- a/third_party/rust/packed_simd/src/api/math.rs +++ /dev/null @@ -1,4 +0,0 @@ -//! Implements vertical math operations - -#[macro_use] -mod float; diff --git a/third_party/rust/packed_simd/src/api/math/float.rs b/third_party/rust/packed_simd/src/api/math/float.rs deleted file mode 100644 index d5d2bee2eb..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float.rs +++ /dev/null @@ -1,64 +0,0 @@ -//! Implements vertical floating-point math operations. - -#[macro_use] -mod abs; - -#[macro_use] -mod consts; - -#[macro_use] -mod cos; - -#[macro_use] -mod exp; - -#[macro_use] -mod powf; - -#[macro_use] -mod ln; - -#[macro_use] -mod mul_add; - -#[macro_use] -mod mul_adde; - -#[macro_use] -mod recpre; - -#[macro_use] -mod rsqrte; - -#[macro_use] -mod sin; - -#[macro_use] -mod sqrt; - -#[macro_use] -mod sqrte; - -#[macro_use] -mod tanh; - -macro_rules! impl_float_category { - ([$elem_ty:ident; $elem_count:expr]: $id:ident, $mask_ty:ident) => { - impl $id { - #[inline] - pub fn is_nan(self) -> $mask_ty { - self.ne(self) - } - - #[inline] - pub fn is_infinite(self) -> $mask_ty { - self.eq(Self::INFINITY) | self.eq(Self::NEG_INFINITY) - } - - #[inline] - pub fn is_finite(self) -> $mask_ty { - !(self.is_nan() | self.is_infinite()) - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/abs.rs b/third_party/rust/packed_simd/src/api/math/float/abs.rs deleted file mode 100644 index 1865bdb68e..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/abs.rs +++ /dev/null @@ -1,31 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `abs`. - -macro_rules! impl_math_float_abs { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Absolute value. - #[inline] - pub fn abs(self) -> Self { - use crate::codegen::math::float::abs::Abs; - Abs::abs(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_abs>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn abs() { - let o = $id::splat(1 as $elem_ty); - assert_eq!(o, o.abs()); - - let mo = $id::splat(-1 as $elem_ty); - assert_eq!(o, mo.abs()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/consts.rs b/third_party/rust/packed_simd/src/api/math/float/consts.rs deleted file mode 100644 index 7f41acbf1b..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/consts.rs +++ /dev/null @@ -1,74 +0,0 @@ -macro_rules! impl_float_consts { - ([$elem_ty:ident; $elem_count:expr]: $id:ident) => { - impl $id { - /// Machine epsilon value. - pub const EPSILON: $id = $id::splat(core::$elem_ty::EPSILON); - - /// Smallest finite value. - pub const MIN: $id = $id::splat(core::$elem_ty::MIN); - - /// Smallest positive normal value. - pub const MIN_POSITIVE: $id = $id::splat(core::$elem_ty::MIN_POSITIVE); - - /// Largest finite value. - pub const MAX: $id = $id::splat(core::$elem_ty::MAX); - - /// Not a Number (NaN). - pub const NAN: $id = $id::splat(core::$elem_ty::NAN); - - /// Infinity (∞). - pub const INFINITY: $id = $id::splat(core::$elem_ty::INFINITY); - - /// Negative infinity (-∞). - pub const NEG_INFINITY: $id = $id::splat(core::$elem_ty::NEG_INFINITY); - - /// Archimedes' constant (π) - pub const PI: $id = $id::splat(core::$elem_ty::consts::PI); - - /// π/2 - pub const FRAC_PI_2: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_2); - - /// π/3 - pub const FRAC_PI_3: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_3); - - /// π/4 - pub const FRAC_PI_4: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_4); - - /// π/6 - pub const FRAC_PI_6: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_6); - - /// π/8 - pub const FRAC_PI_8: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_8); - - /// 1/π - pub const FRAC_1_PI: $id = $id::splat(core::$elem_ty::consts::FRAC_1_PI); - - /// 2/π - pub const FRAC_2_PI: $id = $id::splat(core::$elem_ty::consts::FRAC_2_PI); - - /// 2/sqrt(π) - pub const FRAC_2_SQRT_PI: $id = $id::splat(core::$elem_ty::consts::FRAC_2_SQRT_PI); - - /// sqrt(2) - pub const SQRT_2: $id = $id::splat(core::$elem_ty::consts::SQRT_2); - - /// 1/sqrt(2) - pub const FRAC_1_SQRT_2: $id = $id::splat(core::$elem_ty::consts::FRAC_1_SQRT_2); - - /// Euler's number (e) - pub const E: $id = $id::splat(core::$elem_ty::consts::E); - - /// log<sub>2</sub>(e) - pub const LOG2_E: $id = $id::splat(core::$elem_ty::consts::LOG2_E); - - /// log<sub>10</sub>(e) - pub const LOG10_E: $id = $id::splat(core::$elem_ty::consts::LOG10_E); - - /// ln(2) - pub const LN_2: $id = $id::splat(core::$elem_ty::consts::LN_2); - - /// ln(10) - pub const LN_10: $id = $id::splat(core::$elem_ty::consts::LN_10); - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/cos.rs b/third_party/rust/packed_simd/src/api/math/float/cos.rs deleted file mode 100644 index e5b8f46036..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/cos.rs +++ /dev/null @@ -1,44 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `cos`. - -macro_rules! impl_math_float_cos { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Cosine. - #[inline] - pub fn cos(self) -> Self { - use crate::codegen::math::float::cos::Cos; - Cos::cos(self) - } - - /// Cosine of `self * PI`. - #[inline] - pub fn cos_pi(self) -> Self { - use crate::codegen::math::float::cos_pi::CosPi; - CosPi::cos_pi(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_cos>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn cos() { - use crate::$elem_ty::consts::PI; - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let p = $id::splat(PI as $elem_ty); - let ph = $id::splat(PI as $elem_ty / 2.); - let z_r = $id::splat((PI as $elem_ty / 2.).cos()); - let o_r = $id::splat((PI as $elem_ty).cos()); - - assert_eq!(o, z.cos()); - assert_eq!(z_r, ph.cos()); - assert_eq!(o_r, p.cos()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/exp.rs b/third_party/rust/packed_simd/src/api/math/float/exp.rs deleted file mode 100644 index e3356d853a..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/exp.rs +++ /dev/null @@ -1,33 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `exp`. - -macro_rules! impl_math_float_exp { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Returns the exponential function of `self`: `e^(self)`. - #[inline] - pub fn exp(self) -> Self { - use crate::codegen::math::float::exp::Exp; - Exp::exp(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_exp>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn exp() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - assert_eq!(o, z.exp()); - - let e = $id::splat(crate::f64::consts::E as $elem_ty); - let tol = $id::splat(2.4e-4 as $elem_ty); - assert!((e - o.exp()).abs().le(tol).all()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/ln.rs b/third_party/rust/packed_simd/src/api/math/float/ln.rs deleted file mode 100644 index 5ceb9173ae..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/ln.rs +++ /dev/null @@ -1,33 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `ln`. - -macro_rules! impl_math_float_ln { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Returns the natural logarithm of `self`. - #[inline] - pub fn ln(self) -> Self { - use crate::codegen::math::float::ln::Ln; - Ln::ln(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_ln>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ln() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - assert_eq!(z, o.ln()); - - let e = $id::splat(crate::f64::consts::E as $elem_ty); - let tol = $id::splat(2.4e-4 as $elem_ty); - assert!((o - e.ln()).abs().le(tol).all()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/mul_add.rs b/third_party/rust/packed_simd/src/api/math/float/mul_add.rs deleted file mode 100644 index 4b170ee2b7..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/mul_add.rs +++ /dev/null @@ -1,44 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `mul_add`. - -macro_rules! impl_math_float_mul_add { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Fused multiply add: `self * y + z` - #[inline] - pub fn mul_add(self, y: Self, z: Self) -> Self { - use crate::codegen::math::float::mul_add::MulAdd; - MulAdd::mul_add(self, y, z) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_mul_add>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn mul_add() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let t3 = $id::splat(3 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - assert_eq!(z, z.mul_add(z, z)); - assert_eq!(o, o.mul_add(o, z)); - assert_eq!(o, o.mul_add(z, o)); - assert_eq!(o, z.mul_add(o, o)); - - assert_eq!(t, o.mul_add(o, o)); - assert_eq!(t, o.mul_add(t, z)); - assert_eq!(t, t.mul_add(o, z)); - - assert_eq!(f, t.mul_add(t, z)); - assert_eq!(f, t.mul_add(o, t)); - assert_eq!(t3, t.mul_add(o, o)); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/mul_adde.rs b/third_party/rust/packed_simd/src/api/math/float/mul_adde.rs deleted file mode 100644 index c5b27110f2..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/mul_adde.rs +++ /dev/null @@ -1,48 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `mul_adde`. - -macro_rules! impl_math_float_mul_adde { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Fused multiply add estimate: ~= `self * y + z` - /// - /// While fused multiply-add (`fma`) has infinite precision, - /// `mul_adde` has _at worst_ the same precision of a multiply followed by an add. - /// This might be more efficient on architectures that do not have an `fma` instruction. - #[inline] - pub fn mul_adde(self, y: Self, z: Self) -> Self { - use crate::codegen::math::float::mul_adde::MulAddE; - MulAddE::mul_adde(self, y, z) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_mul_adde>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn mul_adde() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let t3 = $id::splat(3 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - assert_eq!(z, z.mul_adde(z, z)); - assert_eq!(o, o.mul_adde(o, z)); - assert_eq!(o, o.mul_adde(z, o)); - assert_eq!(o, z.mul_adde(o, o)); - - assert_eq!(t, o.mul_adde(o, o)); - assert_eq!(t, o.mul_adde(t, z)); - assert_eq!(t, t.mul_adde(o, z)); - - assert_eq!(f, t.mul_adde(t, z)); - assert_eq!(f, t.mul_adde(o, t)); - assert_eq!(t3, t.mul_adde(o, o)); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/powf.rs b/third_party/rust/packed_simd/src/api/math/float/powf.rs deleted file mode 100644 index 83dc9ff9c0..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/powf.rs +++ /dev/null @@ -1,36 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `powf`. - -macro_rules! impl_math_float_powf { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Raises `self` number to the floating point power of `x`. - #[inline] - pub fn powf(self, x: Self) -> Self { - use crate::codegen::math::float::powf::Powf; - Powf::powf(self, x) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_powf>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn powf() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - assert_eq!(o, o.powf(z)); - assert_eq!(o, t.powf(z)); - assert_eq!(o, o.powf(o)); - assert_eq!(t, t.powf(o)); - - let f = $id::splat(4 as $elem_ty); - assert_eq!(f, t.powf(t)); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/recpre.rs b/third_party/rust/packed_simd/src/api/math/float/recpre.rs deleted file mode 100644 index 127f0b2ff6..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/recpre.rs +++ /dev/null @@ -1,36 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `recpre`. - -macro_rules! impl_math_float_recpre { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Reciprocal estimate: `~= 1. / self`. - /// - /// FIXME: The precision of the estimate is currently unspecified. - #[inline] - pub fn recpre(self) -> Self { - $id::splat(1.) / self - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_recpre>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn recpre() { - let tol = $id::splat(2.4e-4 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let error = (o - o.recpre()).abs(); - assert!(error.le(tol).all()); - - let t = $id::splat(2 as $elem_ty); - let e = 0.5; - let error = (e - t.recpre()).abs(); - assert!(error.le(tol).all()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/rsqrte.rs b/third_party/rust/packed_simd/src/api/math/float/rsqrte.rs deleted file mode 100644 index c77977f7b1..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/rsqrte.rs +++ /dev/null @@ -1,40 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `rsqrte`. - -macro_rules! impl_math_float_rsqrte { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Reciprocal square-root estimate: `~= 1. / self.sqrt()`. - /// - /// FIXME: The precision of the estimate is currently unspecified. - #[inline] - pub fn rsqrte(self) -> Self { - unsafe { - use crate::llvm::simd_fsqrt; - $id::splat(1.) / Simd(simd_fsqrt(self.0)) - } - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_rsqrte>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn rsqrte() { - use crate::$elem_ty::consts::SQRT_2; - let tol = $id::splat(2.4e-4 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let error = (o - o.rsqrte()).abs(); - assert!(error.le(tol).all()); - - let t = $id::splat(2 as $elem_ty); - let e = 1. / SQRT_2; - let error = (e - t.rsqrte()).abs(); - assert!(error.le(tol).all()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/sin.rs b/third_party/rust/packed_simd/src/api/math/float/sin.rs deleted file mode 100644 index 49908319b1..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/sin.rs +++ /dev/null @@ -1,50 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `sin`. - -macro_rules! impl_math_float_sin { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Sine. - #[inline] - pub fn sin(self) -> Self { - use crate::codegen::math::float::sin::Sin; - Sin::sin(self) - } - - /// Sine of `self * PI`. - #[inline] - pub fn sin_pi(self) -> Self { - use crate::codegen::math::float::sin_pi::SinPi; - SinPi::sin_pi(self) - } - - /// Sine and cosine of `self * PI`. - #[inline] - pub fn sin_cos_pi(self) -> (Self, Self) { - use crate::codegen::math::float::sin_cos_pi::SinCosPi; - SinCosPi::sin_cos_pi(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_sin>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn sin() { - use crate::$elem_ty::consts::PI; - let z = $id::splat(0 as $elem_ty); - let p = $id::splat(PI as $elem_ty); - let ph = $id::splat(PI as $elem_ty / 2.); - let o_r = $id::splat((PI as $elem_ty / 2.).sin()); - let z_r = $id::splat((PI as $elem_ty).sin()); - - assert_eq!(z, z.sin()); - assert_eq!(o_r, ph.sin()); - assert_eq!(z_r, p.sin()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/sqrt.rs b/third_party/rust/packed_simd/src/api/math/float/sqrt.rs deleted file mode 100644 index ae624122d0..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/sqrt.rs +++ /dev/null @@ -1,35 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `sqrt`. - -macro_rules! impl_math_float_sqrt { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - #[inline] - pub fn sqrt(self) -> Self { - use crate::codegen::math::float::sqrt::Sqrt; - Sqrt::sqrt(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_sqrt>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn sqrt() { - use crate::$elem_ty::consts::SQRT_2; - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - assert_eq!(z, z.sqrt()); - assert_eq!(o, o.sqrt()); - - let t = $id::splat(2 as $elem_ty); - let e = $id::splat(SQRT_2); - assert_eq!(e, t.sqrt()); - - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/sqrte.rs b/third_party/rust/packed_simd/src/api/math/float/sqrte.rs deleted file mode 100644 index f7ffad748d..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/sqrte.rs +++ /dev/null @@ -1,44 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `sqrte`. - -macro_rules! impl_math_float_sqrte { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Square-root estimate. - /// - /// FIXME: The precision of the estimate is currently unspecified. - #[inline] - pub fn sqrte(self) -> Self { - use crate::codegen::math::float::sqrte::Sqrte; - Sqrte::sqrte(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_sqrte>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn sqrte() { - use crate::$elem_ty::consts::SQRT_2; - let tol = $id::splat(2.4e-4 as $elem_ty); - - let z = $id::splat(0 as $elem_ty); - let error = (z - z.sqrte()).abs(); - assert!(error.le(tol).all()); - - let o = $id::splat(1 as $elem_ty); - let error = (o - o.sqrte()).abs(); - assert!(error.le(tol).all()); - - let t = $id::splat(2 as $elem_ty); - let e = $id::splat(SQRT_2 as $elem_ty); - let error = (e - t.sqrte()).abs(); - - assert!(error.le(tol).all()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/math/float/tanh.rs b/third_party/rust/packed_simd/src/api/math/float/tanh.rs deleted file mode 100644 index acfd93caaa..0000000000 --- a/third_party/rust/packed_simd/src/api/math/float/tanh.rs +++ /dev/null @@ -1,29 +0,0 @@ -//! Implements vertical (lane-wise) floating-point `tanh`. - -macro_rules! impl_math_float_tanh { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Tanh. - #[inline] - pub fn tanh(self) -> Self { - use crate::codegen::math::float::tanh::Tanh; - Tanh::tanh(self) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _math_tanh>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn tanh() { - let z = $id::splat(0 as $elem_ty); - - assert_eq!(z, z.tanh()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/minimal.rs b/third_party/rust/packed_simd/src/api/minimal.rs deleted file mode 100644 index 840d9e3258..0000000000 --- a/third_party/rust/packed_simd/src/api/minimal.rs +++ /dev/null @@ -1,6 +0,0 @@ -#[macro_use] -mod iuf; -#[macro_use] -mod mask; -#[macro_use] -mod ptr; diff --git a/third_party/rust/packed_simd/src/api/minimal/iuf.rs b/third_party/rust/packed_simd/src/api/minimal/iuf.rs deleted file mode 100644 index a155ac178a..0000000000 --- a/third_party/rust/packed_simd/src/api/minimal/iuf.rs +++ /dev/null @@ -1,169 +0,0 @@ -//! Minimal API of signed integer, unsigned integer, and floating-point -//! vectors. - -macro_rules! impl_minimal_iuf { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $ielem_ty:ident | - $test_tt:tt | $($elem_name:ident),+ | $(#[$doc:meta])*) => { - - $(#[$doc])* - pub type $id = Simd<[$elem_ty; $elem_count]>; - - impl sealed::Simd for $id { - type Element = $elem_ty; - const LANES: usize = $elem_count; - type LanesType = [u32; $elem_count]; - } - - impl $id { - /// Creates a new instance with each vector elements initialized - /// with the provided values. - #[inline] - #[allow(clippy::too_many_arguments)] - pub const fn new($($elem_name: $elem_ty),*) -> Self { - Simd(codegen::$id($($elem_name as $ielem_ty),*)) - } - - /// Returns the number of vector lanes. - #[inline] - pub const fn lanes() -> usize { - $elem_count - } - - /// Constructs a new instance with each element initialized to - /// `value`. - #[inline] - pub const fn splat(value: $elem_ty) -> Self { - Simd(codegen::$id($({ - #[allow(non_camel_case_types, dead_code)] - struct $elem_name; - value as $ielem_ty - }),*)) - } - - /// Extracts the value at `index`. - /// - /// # Panics - /// - /// If `index >= Self::lanes()`. - #[inline] - pub fn extract(self, index: usize) -> $elem_ty { - assert!(index < $elem_count); - unsafe { self.extract_unchecked(index) } - } - - /// Extracts the value at `index`. - /// - /// # Safety - /// - /// If `index >= Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn extract_unchecked(self, index: usize) -> $elem_ty { - use crate::llvm::simd_extract; - let e: $ielem_ty = simd_extract(self.0, index as u32); - e as $elem_ty - } - - /// Returns a new vector where the value at `index` is replaced by `new_value`. - /// - /// # Panics - /// - /// If `index >= Self::lanes()`. - #[inline] - #[must_use = "replace does not modify the original value - \ - it returns a new vector with the value at `index` \ - replaced by `new_value`d" - ] - pub fn replace(self, index: usize, new_value: $elem_ty) -> Self { - assert!(index < $elem_count); - unsafe { self.replace_unchecked(index, new_value) } - } - - /// Returns a new vector where the value at `index` is replaced by `new_value`. - /// - /// # Safety - /// - /// If `index >= Self::lanes()` the behavior is undefined. - #[inline] - #[must_use = "replace_unchecked does not modify the original value - \ - it returns a new vector with the value at `index` \ - replaced by `new_value`d" - ] - pub unsafe fn replace_unchecked( - self, - index: usize, - new_value: $elem_ty, - ) -> Self { - use crate::llvm::simd_insert; - Simd(simd_insert(self.0, index as u32, new_value as $ielem_ty)) - } - } - - test_if!{ - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _minimal>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn minimal() { - // lanes: - assert_eq!($elem_count, $id::lanes()); - - // splat and extract / extract_unchecked: - const VAL: $elem_ty = 7 as $elem_ty; - const VEC: $id = $id::splat(VAL); - for i in 0..$id::lanes() { - assert_eq!(VAL, VEC.extract(i)); - assert_eq!( - VAL, unsafe { VEC.extract_unchecked(i) } - ); - } - - // replace / replace_unchecked - let new_vec = VEC.replace(0, 42 as $elem_ty); - for i in 0..$id::lanes() { - if i == 0 { - assert_eq!(42 as $elem_ty, new_vec.extract(i)); - } else { - assert_eq!(VAL, new_vec.extract(i)); - } - } - let new_vec = unsafe { - VEC.replace_unchecked(0, 42 as $elem_ty) - }; - for i in 0..$id::lanes() { - if i == 0 { - assert_eq!(42 as $elem_ty, new_vec.extract(i)); - } else { - assert_eq!(VAL, new_vec.extract(i)); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn extract_panic_oob() { - const VAL: $elem_ty = 7 as $elem_ty; - const VEC: $id = $id::splat(VAL); - let _ = VEC.extract($id::lanes()); - } - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn replace_panic_oob() { - const VAL: $elem_ty = 7 as $elem_ty; - const VEC: $id = $id::splat(VAL); - let _ = VEC.replace($id::lanes(), 42 as $elem_ty); - } - } - } - } - } -} diff --git a/third_party/rust/packed_simd/src/api/minimal/mask.rs b/third_party/rust/packed_simd/src/api/minimal/mask.rs deleted file mode 100644 index a420060b42..0000000000 --- a/third_party/rust/packed_simd/src/api/minimal/mask.rs +++ /dev/null @@ -1,176 +0,0 @@ -//! Minimal API of mask vectors. - -macro_rules! impl_minimal_mask { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $ielem_ty:ident - | $test_tt:tt | $($elem_name:ident),+ | $(#[$doc:meta])*) => { - $(#[$doc])* - pub type $id = Simd<[$elem_ty; $elem_count]>; - - impl sealed::Simd for $id { - type Element = $elem_ty; - const LANES: usize = $elem_count; - type LanesType = [u32; $elem_count]; - } - - impl $id { - /// Creates a new instance with each vector elements initialized - /// with the provided values. - #[inline] - #[allow(clippy::too_many_arguments)] - pub const fn new($($elem_name: bool),*) -> Self { - Simd(codegen::$id($(Self::bool_to_internal($elem_name)),*)) - } - - /// Converts a boolean type into the type of the vector lanes. - #[inline] - #[allow(clippy::indexing_slicing)] - const fn bool_to_internal(x: bool) -> $ielem_ty { - [0 as $ielem_ty, !(0 as $ielem_ty)][x as usize] - } - - /// Returns the number of vector lanes. - #[inline] - pub const fn lanes() -> usize { - $elem_count - } - - /// Constructs a new instance with each element initialized to - /// `value`. - #[inline] - pub const fn splat(value: bool) -> Self { - Simd(codegen::$id($({ - #[allow(non_camel_case_types, dead_code)] - struct $elem_name; - Self::bool_to_internal(value) - }),*)) - } - - /// Extracts the value at `index`. - /// - /// # Panics - /// - /// If `index >= Self::lanes()`. - #[inline] - pub fn extract(self, index: usize) -> bool { - assert!(index < $elem_count); - unsafe { self.extract_unchecked(index) } - } - - /// Extracts the value at `index`. - /// - /// # Safety - /// - /// If `index >= Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn extract_unchecked(self, index: usize) -> bool { - use crate::llvm::simd_extract; - let x: $ielem_ty = simd_extract(self.0, index as u32); - x != 0 - } - - /// Returns a new vector where the value at `index` is replaced by - /// `new_value`. - /// - /// # Panics - /// - /// If `index >= Self::lanes()`. - #[inline] - #[must_use = "replace does not modify the original value - \ - it returns a new vector with the value at `index` \ - replaced by `new_value`d" - ] - pub fn replace(self, index: usize, new_value: bool) -> Self { - assert!(index < $elem_count); - unsafe { self.replace_unchecked(index, new_value) } - } - - /// Returns a new vector where the value at `index` is replaced by - /// `new_value`. - /// - /// # Safety - /// - /// If `index >= Self::lanes()` the behavior is undefined. - #[inline] - #[must_use = "replace_unchecked does not modify the original value - \ - it returns a new vector with the value at `index` \ - replaced by `new_value`d" - ] - pub unsafe fn replace_unchecked( - self, - index: usize, - new_value: bool, - ) -> Self { - use crate::llvm::simd_insert; - Simd(simd_insert(self.0, index as u32, - Self::bool_to_internal(new_value))) - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _minimal>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn minimal() { - // TODO: test new - - // lanes: - assert_eq!($elem_count, $id::lanes()); - - // splat and extract / extract_unchecked: - let vec = $id::splat(true); - for i in 0..$id::lanes() { - assert_eq!(true, vec.extract(i)); - assert_eq!(true, - unsafe { vec.extract_unchecked(i) } - ); - } - - // replace / replace_unchecked - let new_vec = vec.replace(0, false); - for i in 0..$id::lanes() { - if i == 0 { - assert_eq!(false, new_vec.extract(i)); - } else { - assert_eq!(true, new_vec.extract(i)); - } - } - let new_vec = unsafe { - vec.replace_unchecked(0, false) - }; - for i in 0..$id::lanes() { - if i == 0 { - assert_eq!(false, new_vec.extract(i)); - } else { - assert_eq!(true, new_vec.extract(i)); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn extract_panic_oob() { - let vec = $id::splat(false); - let _ = vec.extract($id::lanes()); - } - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn replace_panic_oob() { - let vec = $id::splat(false); - let _ = vec.replace($id::lanes(), true); - } - } - } - } - } -} diff --git a/third_party/rust/packed_simd/src/api/minimal/ptr.rs b/third_party/rust/packed_simd/src/api/minimal/ptr.rs deleted file mode 100644 index d9e47c9ccb..0000000000 --- a/third_party/rust/packed_simd/src/api/minimal/ptr.rs +++ /dev/null @@ -1,1373 +0,0 @@ -//! Minimal API of pointer vectors. - -macro_rules! impl_minimal_p { - ([$elem_ty:ty; $elem_count:expr]: $id:ident, $mask_ty:ident, - $usize_ty:ident, $isize_ty:ident | $ref:ident | $test_tt:tt - | $($elem_name:ident),+ | ($true:expr, $false:expr) | - $(#[$doc:meta])*) => { - - $(#[$doc])* - pub type $id<T> = Simd<[$elem_ty; $elem_count]>; - - impl<T> sealed::Simd for $id<T> { - type Element = $elem_ty; - const LANES: usize = $elem_count; - type LanesType = [u32; $elem_count]; - } - - impl<T> $id<T> { - /// Creates a new instance with each vector elements initialized - /// with the provided values. - #[inline] - #[allow(clippy::too_many_arguments)] - pub const fn new($($elem_name: $elem_ty),*) -> Self { - Simd(codegen::$id($($elem_name),*)) - } - - /// Returns the number of vector lanes. - #[inline] - pub const fn lanes() -> usize { - $elem_count - } - - /// Constructs a new instance with each element initialized to - /// `value`. - #[inline] - pub const fn splat(value: $elem_ty) -> Self { - Simd(codegen::$id($({ - #[allow(non_camel_case_types, dead_code)] - struct $elem_name; - value - }),*)) - } - - /// Constructs a new instance with each element initialized to - /// `null`. - #[inline] - pub const fn null() -> Self { - Self::splat(crate::ptr::null_mut() as $elem_ty) - } - - /// Returns a mask that selects those lanes that contain `null` - /// pointers. - #[inline] - pub fn is_null(self) -> $mask_ty { - self.eq(Self::null()) - } - - /// Extracts the value at `index`. - /// - /// # Panics - /// - /// If `index >= Self::lanes()`. - #[inline] - pub fn extract(self, index: usize) -> $elem_ty { - assert!(index < $elem_count); - unsafe { self.extract_unchecked(index) } - } - - /// Extracts the value at `index`. - /// - /// # Safety - /// - /// If `index >= Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn extract_unchecked(self, index: usize) -> $elem_ty { - use crate::llvm::simd_extract; - simd_extract(self.0, index as u32) - } - - /// Returns a new vector where the value at `index` is replaced by - /// `new_value`. - /// - /// # Panics - /// - /// If `index >= Self::lanes()`. - #[inline] - #[must_use = "replace does not modify the original value - \ - it returns a new vector with the value at `index` \ - replaced by `new_value`d" - ] - #[allow(clippy::not_unsafe_ptr_arg_deref)] - pub fn replace(self, index: usize, new_value: $elem_ty) -> Self { - assert!(index < $elem_count); - unsafe { self.replace_unchecked(index, new_value) } - } - - /// Returns a new vector where the value at `index` is replaced by `new_value`. - /// - /// # Safety - /// - /// If `index >= Self::lanes()` the behavior is undefined. - #[inline] - #[must_use = "replace_unchecked does not modify the original value - \ - it returns a new vector with the value at `index` \ - replaced by `new_value`d" - ] - pub unsafe fn replace_unchecked( - self, - index: usize, - new_value: $elem_ty, - ) -> Self { - use crate::llvm::simd_insert; - Simd(simd_insert(self.0, index as u32, new_value)) - } - } - - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _minimal>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn minimal() { - // lanes: - assert_eq!($elem_count, $id::<i32>::lanes()); - - // splat and extract / extract_unchecked: - let VAL7: <$id<i32> as sealed::Simd>::Element - = $ref!(7); - let VAL42: <$id<i32> as sealed::Simd>::Element - = $ref!(42); - let VEC: $id<i32> = $id::splat(VAL7); - for i in 0..$id::<i32>::lanes() { - assert_eq!(VAL7, VEC.extract(i)); - assert_eq!( - VAL7, unsafe { VEC.extract_unchecked(i) } - ); - } - - // replace / replace_unchecked - let new_vec = VEC.replace(0, VAL42); - for i in 0..$id::<i32>::lanes() { - if i == 0 { - assert_eq!(VAL42, new_vec.extract(i)); - } else { - assert_eq!(VAL7, new_vec.extract(i)); - } - } - let new_vec = unsafe { - VEC.replace_unchecked(0, VAL42) - }; - for i in 0..$id::<i32>::lanes() { - if i == 0 { - assert_eq!(VAL42, new_vec.extract(i)); - } else { - assert_eq!(VAL7, new_vec.extract(i)); - } - } - - let mut n = $id::<i32>::null(); - assert_eq!( - n, - $id::<i32>::splat(unsafe { crate::mem::zeroed() }) - ); - assert!(n.is_null().all()); - n = n.replace( - 0, unsafe { crate::mem::transmute(1_isize) } - ); - assert!(!n.is_null().all()); - if $id::<i32>::lanes() > 1 { - assert!(n.is_null().any()); - } else { - assert!(!n.is_null().any()); - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn extract_panic_oob() { - let VAL: <$id<i32> as sealed::Simd>::Element - = $ref!(7); - let VEC: $id<i32> = $id::splat(VAL); - let _ = VEC.extract($id::<i32>::lanes()); - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn replace_panic_oob() { - let VAL: <$id<i32> as sealed::Simd>::Element - = $ref!(7); - let VAL42: <$id<i32> as sealed::Simd>::Element - = $ref!(42); - let VEC: $id<i32> = $id::splat(VAL); - let _ = VEC.replace($id::<i32>::lanes(), VAL42); - } - } - } - } - - impl<T> crate::fmt::Debug for $id<T> { - #[allow(clippy::missing_inline_in_public_items)] - fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) - -> crate::fmt::Result { - write!( - f, - "{}<{}>(", - stringify!($id), - crate::intrinsics::type_name::<T>() - )?; - for i in 0..$elem_count { - if i > 0 { - write!(f, ", ")?; - } - self.extract(i).fmt(f)?; - } - write!(f, ")") - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _fmt_debug>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn debug() { - use arrayvec::{ArrayString,ArrayVec}; - type TinyString = ArrayString<[u8; 512]>; - - use crate::fmt::Write; - let v = $id::<i32>::default(); - let mut s = TinyString::new(); - write!(&mut s, "{:?}", v).unwrap(); - - let mut beg = TinyString::new(); - write!(&mut beg, "{}<i32>(", stringify!($id)).unwrap(); - assert!( - s.starts_with(beg.as_str()), - "s = {} (should start with = {})", s, beg - ); - assert!(s.ends_with(")")); - let s: ArrayVec<[TinyString; 64]> - = s.replace(beg.as_str(), "") - .replace(")", "").split(",") - .map(|v| TinyString::from(v.trim()).unwrap()) - .collect(); - assert_eq!(s.len(), $id::<i32>::lanes()); - for (index, ss) in s.into_iter().enumerate() { - let mut e = TinyString::new(); - write!(&mut e, "{:?}", v.extract(index)).unwrap(); - assert_eq!(ss, e); - } - } - } - } - } - - impl<T> Default for $id<T> { - #[inline] - fn default() -> Self { - // FIXME: ptrs do not implement default - Self::null() - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _default>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn default() { - let a = $id::<i32>::default(); - for i in 0..$id::<i32>::lanes() { - assert_eq!( - a.extract(i), unsafe { crate::mem::zeroed() } - ); - } - } - } - } - } - - impl<T> $id<T> { - /// Lane-wise equality comparison. - #[inline] - pub fn eq(self, other: Self) -> $mask_ty { - unsafe { - use crate::llvm::simd_eq; - let a: $usize_ty = crate::mem::transmute(self); - let b: $usize_ty = crate::mem::transmute(other); - Simd(simd_eq(a.0, b.0)) - } - } - - /// Lane-wise inequality comparison. - #[inline] - pub fn ne(self, other: Self) -> $mask_ty { - unsafe { - use crate::llvm::simd_ne; - let a: $usize_ty = crate::mem::transmute(self); - let b: $usize_ty = crate::mem::transmute(other); - Simd(simd_ne(a.0, b.0)) - } - } - - /// Lane-wise less-than comparison. - #[inline] - pub fn lt(self, other: Self) -> $mask_ty { - unsafe { - use crate::llvm::simd_lt; - let a: $usize_ty = crate::mem::transmute(self); - let b: $usize_ty = crate::mem::transmute(other); - Simd(simd_lt(a.0, b.0)) - } - } - - /// Lane-wise less-than-or-equals comparison. - #[inline] - pub fn le(self, other: Self) -> $mask_ty { - unsafe { - use crate::llvm::simd_le; - let a: $usize_ty = crate::mem::transmute(self); - let b: $usize_ty = crate::mem::transmute(other); - Simd(simd_le(a.0, b.0)) - } - } - - /// Lane-wise greater-than comparison. - #[inline] - pub fn gt(self, other: Self) -> $mask_ty { - unsafe { - use crate::llvm::simd_gt; - let a: $usize_ty = crate::mem::transmute(self); - let b: $usize_ty = crate::mem::transmute(other); - Simd(simd_gt(a.0, b.0)) - } - } - - /// Lane-wise greater-than-or-equals comparison. - #[inline] - pub fn ge(self, other: Self) -> $mask_ty { - unsafe { - use crate::llvm::simd_ge; - let a: $usize_ty = crate::mem::transmute(self); - let b: $usize_ty = crate::mem::transmute(other); - Simd(simd_ge(a.0, b.0)) - } - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_vertical>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn cmp() { - let a = $id::<i32>::null(); - let b = $id::<i32>::splat(unsafe { - crate::mem::transmute(1_isize) - }); - - let r = a.lt(b); - let e = $mask_ty::splat(true); - assert!(r == e); - let r = a.le(b); - assert!(r == e); - - let e = $mask_ty::splat(false); - let r = a.gt(b); - assert!(r == e); - let r = a.ge(b); - assert!(r == e); - let r = a.eq(b); - assert!(r == e); - - let mut a = a; - let mut b = b; - let mut e = e; - for i in 0..$id::<i32>::lanes() { - if i % 2 == 0 { - a = a.replace( - i, - unsafe { crate::mem::transmute(0_isize) } - ); - b = b.replace( - i, - unsafe { crate::mem::transmute(1_isize) } - ); - e = e.replace(i, true); - } else { - a = a.replace( - i, - unsafe { crate::mem::transmute(1_isize) } - ); - b = b.replace( - i, - unsafe { crate::mem::transmute(0_isize) } - ); - e = e.replace(i, false); - } - } - let r = a.lt(b); - assert!(r == e); - } - } - } - } - - #[allow(clippy::partialeq_ne_impl)] - impl<T> crate::cmp::PartialEq<$id<T>> for $id<T> { - #[inline] - fn eq(&self, other: &Self) -> bool { - $id::<T>::eq(*self, *other).all() - } - #[inline] - fn ne(&self, other: &Self) -> bool { - $id::<T>::ne(*self, *other).any() - } - } - - // FIXME: https://github.com/rust-lang-nursery/rust-clippy/issues/2892 - #[allow(clippy::partialeq_ne_impl)] - impl<T> crate::cmp::PartialEq<LexicographicallyOrdered<$id<T>>> - for LexicographicallyOrdered<$id<T>> - { - #[inline] - fn eq(&self, other: &Self) -> bool { - self.0 == other.0 - } - #[inline] - fn ne(&self, other: &Self) -> bool { - self.0 != other.0 - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_PartialEq>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn partial_eq() { - let a = $id::<i32>::null(); - let b = $id::<i32>::splat(unsafe { - crate::mem::transmute(1_isize) - }); - - assert!(a != b); - assert!(!(a == b)); - assert!(a == a); - assert!(!(a != a)); - - if $id::<i32>::lanes() > 1 { - let a = $id::<i32>::null().replace(0, unsafe { - crate::mem::transmute(1_isize) - }); - let b = $id::<i32>::splat(unsafe { - crate::mem::transmute(1_isize) - }); - - assert!(a != b); - assert!(!(a == b)); - assert!(a == a); - assert!(!(a != a)); - } - } - } - } - } - - impl<T> crate::cmp::Eq for $id<T> {} - impl<T> crate::cmp::Eq for LexicographicallyOrdered<$id<T>> {} - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _cmp_eq>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn eq() { - fn foo<E: crate::cmp::Eq>(_: E) {} - let a = $id::<i32>::null(); - foo(a); - } - } - } - } - - impl<T> From<[$elem_ty; $elem_count]> for $id<T> { - #[inline] - fn from(array: [$elem_ty; $elem_count]) -> Self { - unsafe { - // FIXME: unnecessary zeroing; better than UB. - let mut u: Self = crate::mem::zeroed(); - crate::ptr::copy_nonoverlapping( - &array as *const [$elem_ty; $elem_count] as *const u8, - &mut u as *mut Self as *mut u8, - crate::mem::size_of::<Self>() - ); - u - } - } - } - impl<T> Into<[$elem_ty; $elem_count]> for $id<T> { - #[inline] - fn into(self) -> [$elem_ty; $elem_count] { - unsafe { - // FIXME: unnecessary zeroing; better than UB. - let mut u: [$elem_ty; $elem_count] = crate::mem::zeroed(); - crate::ptr::copy_nonoverlapping( - &self as *const $id<T> as *const u8, - &mut u as *mut [$elem_ty; $elem_count] as *mut u8, - crate::mem::size_of::<Self>() - ); - u - } - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _from>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn array() { - let values = [1_i32; $elem_count]; - - let mut vec: $id<i32> = Default::default(); - let mut array = [ - $id::<i32>::null().extract(0); $elem_count - ]; - - for i in 0..$elem_count { - let ptr = &values[i] as *const i32 as *mut i32; - vec = vec.replace(i, ptr); - array[i] = ptr; - } - - // FIXME: there is no impl of From<$id<T>> for [$elem_ty; N] - // let a0 = From::from(vec); - // assert_eq!(a0, array); - #[allow(unused_assignments)] - let mut a1 = array; - a1 = vec.into(); - assert_eq!(a1, array); - - let v0: $id<i32> = From::from(array); - assert_eq!(v0, vec); - let v1: $id<i32> = array.into(); - assert_eq!(v1, vec); - } - } - } - } - - impl<T> $id<T> { - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned - /// to an `align_of::<Self>()` boundary. - #[inline] - pub fn from_slice_aligned(slice: &[$elem_ty]) -> Self { - unsafe { - assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_ptr(); - assert!( - target_ptr.align_offset(crate::mem::align_of::<Self>()) - == 0 - ); - Self::from_slice_aligned_unchecked(slice) - } - } - - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()`. - #[inline] - pub fn from_slice_unaligned(slice: &[$elem_ty]) -> Self { - unsafe { - assert!(slice.len() >= $elem_count); - Self::from_slice_unaligned_unchecked(slice) - } - } - - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned - /// to an `align_of::<Self>()` boundary, the behavior is undefined. - #[inline] - pub unsafe fn from_slice_aligned_unchecked(slice: &[$elem_ty]) - -> Self { - #[allow(clippy::cast_ptr_alignment)] - *(slice.as_ptr().cast()) - } - - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn from_slice_unaligned_unchecked( - slice: &[$elem_ty], - ) -> Self { - use crate::mem::size_of; - let target_ptr = slice.as_ptr().cast(); - let mut x = Self::splat(crate::ptr::null_mut() as $elem_ty); - let self_ptr = &mut x as *mut Self as *mut u8; - crate::ptr::copy_nonoverlapping( - target_ptr, - self_ptr, - size_of::<Self>(), - ); - x - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _slice_from_slice>] { - use super::*; - use crate::iter::Iterator; - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from_slice_unaligned() { - let (null, non_null) = ptr_vals!($id<i32>); - - let mut unaligned = [ - non_null; $id::<i32>::lanes() + 1 - ]; - unaligned[0] = null; - let vec = $id::<i32>::from_slice_unaligned( - &unaligned[1..] - ); - for (index, &b) in unaligned.iter().enumerate() { - if index == 0 { - assert_eq!(b, null); - } else { - assert_eq!(b, non_null); - assert_eq!(b, vec.extract(index - 1)); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn from_slice_unaligned_fail() { - let (_null, non_null) = ptr_vals!($id<i32>); - let unaligned = [non_null; $id::<i32>::lanes() + 1]; - // the slice is not large enough => panic - let _vec = $id::<i32>::from_slice_unaligned( - &unaligned[2..] - ); - } - - union A { - data: [<$id<i32> as sealed::Simd>::Element; - 2 * $id::<i32>::lanes()], - _vec: $id<i32>, - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from_slice_aligned() { - let (null, non_null) = ptr_vals!($id<i32>); - let mut aligned = A { - data: [null; 2 * $id::<i32>::lanes()], - }; - for i in - $id::<i32>::lanes()..(2 * $id::<i32>::lanes()) { - unsafe { - aligned.data[i] = non_null; - } - } - - let vec = unsafe { - $id::<i32>::from_slice_aligned( - &aligned.data[$id::<i32>::lanes()..] - ) - }; - for (index, &b) in unsafe { - aligned.data.iter().enumerate() - } { - if index < $id::<i32>::lanes() { - assert_eq!(b, null); - } else { - assert_eq!(b, non_null); - assert_eq!( - b, vec.extract(index - $id::<i32>::lanes()) - ); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn from_slice_aligned_fail_lanes() { - let (_null, non_null) = ptr_vals!($id<i32>); - let aligned = A { - data: [non_null; 2 * $id::<i32>::lanes()], - }; - // the slice is not large enough => panic - let _vec = unsafe { - $id::<i32>::from_slice_aligned( - &aligned.data[2 * $id::<i32>::lanes()..] - ) - }; - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn from_slice_aligned_fail_align() { - unsafe { - let (null, _non_null) = ptr_vals!($id<i32>); - let aligned = A { - data: [null; 2 * $id::<i32>::lanes()], - }; - - // get a pointer to the front of data - let ptr = aligned.data.as_ptr(); - // offset pointer by one element - let ptr = ptr.wrapping_add(1); - - if ptr.align_offset( - crate::mem::align_of::<$id<i32>>() - ) == 0 { - // the pointer is properly aligned, so - // from_slice_aligned won't fail here (e.g. this - // can happen for i128x1). So we panic to make - // the "should_fail" test pass: - panic!("ok"); - } - - // create a slice - this is safe, because the - // elements of the slice exist, are properly - // initialized, and properly aligned: - let s = slice::from_raw_parts( - ptr, $id::<i32>::lanes() - ); - // this should always panic because the slice - // alignment does not match the alignment - // requirements for the vector type: - let _vec = $id::<i32>::from_slice_aligned(s); - } - } - } - } - } - - impl<T> $id<T> { - /// Writes the values of the vector to the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not - /// aligned to an `align_of::<Self>()` boundary. - #[inline] - pub fn write_to_slice_aligned(self, slice: &mut [$elem_ty]) { - unsafe { - assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_mut_ptr(); - assert!( - target_ptr.align_offset(crate::mem::align_of::<Self>()) - == 0 - ); - self.write_to_slice_aligned_unchecked(slice); - } - } - - /// Writes the values of the vector to the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()`. - #[inline] - pub fn write_to_slice_unaligned(self, slice: &mut [$elem_ty]) { - unsafe { - assert!(slice.len() >= $elem_count); - self.write_to_slice_unaligned_unchecked(slice); - } - } - - /// Writes the values of the vector to the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not - /// aligned to an `align_of::<Self>()` boundary, the behavior is - /// undefined. - #[inline] - pub unsafe fn write_to_slice_aligned_unchecked( - self, slice: &mut [$elem_ty], - ) { - #[allow(clippy::cast_ptr_alignment)] - *(slice.as_mut_ptr().cast()) = self; - } - - /// Writes the values of the vector to the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn write_to_slice_unaligned_unchecked( - self, slice: &mut [$elem_ty], - ) { - let target_ptr = slice.as_mut_ptr().cast(); - let self_ptr = &self as *const Self as *const u8; - crate::ptr::copy_nonoverlapping( - self_ptr, - target_ptr, - crate::mem::size_of::<Self>(), - ); - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _slice_write_to_slice>] { - use super::*; - use crate::iter::Iterator; - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn write_to_slice_unaligned() { - let (null, non_null) = ptr_vals!($id<i32>); - let mut unaligned = [null; $id::<i32>::lanes() + 1]; - let vec = $id::<i32>::splat(non_null); - vec.write_to_slice_unaligned(&mut unaligned[1..]); - for (index, &b) in unaligned.iter().enumerate() { - if index == 0 { - assert_eq!(b, null); - } else { - assert_eq!(b, non_null); - assert_eq!(b, vec.extract(index - 1)); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn write_to_slice_unaligned_fail() { - let (null, non_null) = ptr_vals!($id<i32>); - let mut unaligned = [null; $id::<i32>::lanes() + 1]; - let vec = $id::<i32>::splat(non_null); - // the slice is not large enough => panic - vec.write_to_slice_unaligned(&mut unaligned[2..]); - } - - union A { - data: [<$id<i32> as sealed::Simd>::Element; - 2 * $id::<i32>::lanes()], - _vec: $id<i32>, - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn write_to_slice_aligned() { - let (null, non_null) = ptr_vals!($id<i32>); - let mut aligned = A { - data: [null; 2 * $id::<i32>::lanes()], - }; - let vec = $id::<i32>::splat(non_null); - unsafe { - vec.write_to_slice_aligned( - &mut aligned.data[$id::<i32>::lanes()..] - ) - }; - for (index, &b) in - unsafe { aligned.data.iter().enumerate() } { - if index < $id::<i32>::lanes() { - assert_eq!(b, null); - } else { - assert_eq!(b, non_null); - assert_eq!( - b, vec.extract(index - $id::<i32>::lanes()) - ); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn write_to_slice_aligned_fail_lanes() { - let (null, non_null) = ptr_vals!($id<i32>); - let mut aligned = A { - data: [null; 2 * $id::<i32>::lanes()], - }; - let vec = $id::<i32>::splat(non_null); - // the slice is not large enough => panic - unsafe { - vec.write_to_slice_aligned( - &mut aligned.data[2 * $id::<i32>::lanes()..] - ) - }; - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn write_to_slice_aligned_fail_align() { - let (null, non_null) = ptr_vals!($id<i32>); - unsafe { - let mut aligned = A { - data: [null; 2 * $id::<i32>::lanes()], - }; - - // get a pointer to the front of data - let ptr = aligned.data.as_mut_ptr(); - // offset pointer by one element - let ptr = ptr.wrapping_add(1); - - if ptr.align_offset( - crate::mem::align_of::<$id<i32>>() - ) == 0 { - // the pointer is properly aligned, so - // write_to_slice_aligned won't fail here (e.g. - // this can happen for i128x1). So we panic to - // make the "should_fail" test pass: - panic!("ok"); - } - - // create a slice - this is safe, because the - // elements of the slice exist, are properly - // initialized, and properly aligned: - let s = slice::from_raw_parts_mut( - ptr, $id::<i32>::lanes() - ); - // this should always panic because the slice - // alignment does not match the alignment - // requirements for the vector type: - let vec = $id::<i32>::splat(non_null); - vec.write_to_slice_aligned(s); - } - } - } - } - } - - impl<T> crate::hash::Hash for $id<T> { - #[inline] - fn hash<H: crate::hash::Hasher>(&self, state: &mut H) { - let s: $usize_ty = unsafe { crate::mem::transmute(*self) }; - s.hash(state) - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _hash>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn hash() { - use crate::hash::{Hash, Hasher}; - #[allow(deprecated)] - use crate::hash::{SipHasher13}; - - let values = [1_i32; $elem_count]; - - let mut vec: $id<i32> = Default::default(); - let mut array = [ - $id::<i32>::null().extract(0); - $elem_count - ]; - - for i in 0..$elem_count { - let ptr = &values[i] as *const i32 as *mut i32; - vec = vec.replace(i, ptr); - array[i] = ptr; - } - - #[allow(deprecated)] - let mut a_hash = SipHasher13::new(); - let mut v_hash = a_hash.clone(); - array.hash(&mut a_hash); - vec.hash(&mut v_hash); - assert_eq!(a_hash.finish(), v_hash.finish()); - } - } - } - } - - impl<T> $id<T> { - /// Calculates the offset from a pointer. - /// - /// `count` is in units of `T`; e.g. a count of `3` represents a - /// pointer offset of `3 * size_of::<T>()` bytes. - /// - /// # Safety - /// - /// If any of the following conditions are violated, the result is - /// Undefined Behavior: - /// - /// * Both the starting and resulting pointer must be either in - /// bounds or one byte past the end of an allocated object. - /// - /// * The computed offset, in bytes, cannot overflow an `isize`. - /// - /// * The offset being in bounds cannot rely on "wrapping around" - /// the address space. That is, the infinite-precision sum, in bytes - /// must fit in a `usize`. - /// - /// The compiler and standard library generally tries to ensure - /// allocations never reach a size where an offset is a concern. For - /// instance, `Vec` and `Box` ensure they never allocate more than - /// `isize::MAX` bytes, so `vec.as_ptr().offset(vec.len() as isize)` - /// is always safe. - /// - /// Most platforms fundamentally can't even construct such an - /// allocation. For instance, no known 64-bit platform can ever - /// serve a request for 263 bytes due to page-table limitations or - /// splitting the address space. However, some 32-bit and 16-bit - /// platforms may successfully serve a request for more than - /// `isize::MAX` bytes with things like Physical Address Extension. - /// As such, memory acquired directly from allocators or memory - /// mapped files may be too large to handle with this function. - /// - /// Consider using `wrapping_offset` instead if these constraints - /// are difficult to satisfy. The only advantage of this method is - /// that it enables more aggressive compiler optimizations. - #[inline] - pub unsafe fn offset(self, count: $isize_ty) -> Self { - // FIXME: should use LLVM's `add nsw nuw` - self.wrapping_offset(count) - } - - /// Calculates the offset from a pointer using wrapping arithmetic. - /// - /// `count` is in units of `T`; e.g. a count of `3` represents a - /// pointer offset of `3 * size_of::<T>()` bytes. - /// - /// # Safety - /// - /// The resulting pointer does not need to be in bounds, but it is - /// potentially hazardous to dereference (which requires unsafe). - /// - /// Always use `.offset(count)` instead when possible, because - /// offset allows the compiler to optimize better. - #[inline] - pub fn wrapping_offset(self, count: $isize_ty) -> Self { - unsafe { - let x: $isize_ty = crate::mem::transmute(self); - // note: {+,*} currently performs a `wrapping_{add, mul}` - crate::mem::transmute( - x + (count * crate::mem::size_of::<T>() as isize) - ) - } - } - - /// Calculates the distance between two pointers. - /// - /// The returned value is in units of `T`: the distance in bytes is - /// divided by `mem::size_of::<T>()`. - /// - /// This function is the inverse of offset. - /// - /// # Safety - /// - /// If any of the following conditions are violated, the result is - /// Undefined Behavior: - /// - /// * Both the starting and other pointer must be either in bounds - /// or one byte past the end of the same allocated object. - /// - /// * The distance between the pointers, in bytes, cannot overflow - /// an `isize`. - /// - /// * The distance between the pointers, in bytes, must be an exact - /// multiple of the size of `T`. - /// - /// * The distance being in bounds cannot rely on "wrapping around" - /// the address space. - /// - /// The compiler and standard library generally try to ensure - /// allocations never reach a size where an offset is a concern. For - /// instance, `Vec` and `Box` ensure they never allocate more than - /// `isize::MAX` bytes, so `ptr_into_vec.offset_from(vec.as_ptr())` - /// is always safe. - /// - /// Most platforms fundamentally can't even construct such an - /// allocation. For instance, no known 64-bit platform can ever - /// serve a request for 263 bytes due to page-table limitations or - /// splitting the address space. However, some 32-bit and 16-bit - /// platforms may successfully serve a request for more than - /// `isize::MAX` bytes with things like Physical Address Extension. - /// As such, memory acquired directly from allocators or memory - /// mapped files may be too large to handle with this function. - /// - /// Consider using `wrapping_offset_from` instead if these constraints - /// are difficult to satisfy. The only advantage of this method is - /// that it enables more aggressive compiler optimizations. - #[inline] - pub unsafe fn offset_from(self, origin: Self) -> $isize_ty { - // FIXME: should use LLVM's `sub nsw nuw`. - self.wrapping_offset_from(origin) - } - - /// Calculates the distance between two pointers. - /// - /// The returned value is in units of `T`: the distance in bytes is - /// divided by `mem::size_of::<T>()`. - /// - /// If the address different between the two pointers is not a - /// multiple of `mem::size_of::<T>()` then the result of the - /// division is rounded towards zero. - /// - /// Though this method is safe for any two pointers, note that its - /// result will be mostly useless if the two pointers aren't into - /// the same allocated object, for example if they point to two - /// different local variables. - #[inline] - pub fn wrapping_offset_from(self, origin: Self) -> $isize_ty { - let x: $isize_ty = unsafe { crate::mem::transmute(self) }; - let y: $isize_ty = unsafe { crate::mem::transmute(origin) }; - // note: {-,/} currently perform wrapping_{sub, div} - (y - x) / (crate::mem::size_of::<T>() as isize) - } - - /// Calculates the offset from a pointer (convenience for - /// `.offset(count as isize)`). - /// - /// `count` is in units of `T`; e.g. a count of 3 represents a - /// pointer offset of `3 * size_of::<T>()` bytes. - /// - /// # Safety - /// - /// If any of the following conditions are violated, the result is - /// Undefined Behavior: - /// - /// * Both the starting and resulting pointer must be either in - /// bounds or one byte past the end of an allocated object. - /// - /// * The computed offset, in bytes, cannot overflow an `isize`. - /// - /// * The offset being in bounds cannot rely on "wrapping around" - /// the address space. That is, the infinite-precision sum must fit - /// in a `usize`. - /// - /// The compiler and standard library generally tries to ensure - /// allocations never reach a size where an offset is a concern. For - /// instance, `Vec` and `Box` ensure they never allocate more than - /// `isize::MAX` bytes, so `vec.as_ptr().add(vec.len())` is always - /// safe. - /// - /// Most platforms fundamentally can't even construct such an - /// allocation. For instance, no known 64-bit platform can ever - /// serve a request for 263 bytes due to page-table limitations or - /// splitting the address space. However, some 32-bit and 16-bit - /// platforms may successfully serve a request for more than - /// `isize::MAX` bytes with things like Physical Address Extension. - /// As such, memory acquired directly from allocators or memory - /// mapped files may be too large to handle with this function. - /// - /// Consider using `wrapping_offset` instead if these constraints - /// are difficult to satisfy. The only advantage of this method is - /// that it enables more aggressive compiler optimizations. - #[inline] - #[allow(clippy::should_implement_trait)] - pub unsafe fn add(self, count: $usize_ty) -> Self { - self.offset(count.cast()) - } - - /// Calculates the offset from a pointer (convenience for - /// `.offset((count as isize).wrapping_neg())`). - /// - /// `count` is in units of T; e.g. a `count` of 3 represents a - /// pointer offset of `3 * size_of::<T>()` bytes. - /// - /// # Safety - /// - /// If any of the following conditions are violated, the result is - /// Undefined Behavior: - /// - /// * Both the starting and resulting pointer must be either in - /// bounds or one byte past the end of an allocated object. - /// - /// * The computed offset cannot exceed `isize::MAX` **bytes**. - /// - /// * The offset being in bounds cannot rely on "wrapping around" - /// the address space. That is, the infinite-precision sum must fit - /// in a usize. - /// - /// The compiler and standard library generally tries to ensure - /// allocations never reach a size where an offset is a concern. For - /// instance, `Vec` and `Box` ensure they never allocate more than - /// `isize::MAX` bytes, so - /// `vec.as_ptr().add(vec.len()).sub(vec.len())` is always safe. - /// - /// Most platforms fundamentally can't even construct such an - /// allocation. For instance, no known 64-bit platform can ever - /// serve a request for 2<sup>63</sup> bytes due to page-table - /// limitations or splitting the address space. However, some 32-bit - /// and 16-bit platforms may successfully serve a request for more - /// than `isize::MAX` bytes with things like Physical Address - /// Extension. As such, memory acquired directly from allocators or - /// memory mapped files *may* be too large to handle with this - /// function. - /// - /// Consider using `wrapping_offset` instead if these constraints - /// are difficult to satisfy. The only advantage of this method is - /// that it enables more aggressive compiler optimizations. - #[inline] - #[allow(clippy::should_implement_trait)] - pub unsafe fn sub(self, count: $usize_ty) -> Self { - let x: $isize_ty = count.cast(); - // note: - is currently wrapping_neg - self.offset(-x) - } - - /// Calculates the offset from a pointer using wrapping arithmetic. - /// (convenience for `.wrapping_offset(count as isize)`) - /// - /// `count` is in units of T; e.g. a `count` of 3 represents a - /// pointer offset of `3 * size_of::<T>()` bytes. - /// - /// # Safety - /// - /// The resulting pointer does not need to be in bounds, but it is - /// potentially hazardous to dereference (which requires `unsafe`). - /// - /// Always use `.add(count)` instead when possible, because `add` - /// allows the compiler to optimize better. - #[inline] - pub fn wrapping_add(self, count: $usize_ty) -> Self { - self.wrapping_offset(count.cast()) - } - - /// Calculates the offset from a pointer using wrapping arithmetic. - /// (convenience for `.wrapping_offset((count as - /// isize).wrapping_sub())`) - /// - /// `count` is in units of T; e.g. a `count` of 3 represents a - /// pointer offset of `3 * size_of::<T>()` bytes. - /// - /// # Safety - /// - /// The resulting pointer does not need to be in bounds, but it is - /// potentially hazardous to dereference (which requires `unsafe`). - /// - /// Always use `.sub(count)` instead when possible, because `sub` - /// allows the compiler to optimize better. - #[inline] - pub fn wrapping_sub(self, count: $usize_ty) -> Self { - let x: $isize_ty = count.cast(); - self.wrapping_offset(-1 * x) - } - } - - impl<T> $id<T> { - /// Shuffle vector elements according to `indices`. - #[inline] - pub fn shuffle1_dyn<I>(self, indices: I) -> Self - where - Self: codegen::shuffle1_dyn::Shuffle1Dyn<Indices = I>, - { - codegen::shuffle1_dyn::Shuffle1Dyn::shuffle1_dyn(self, indices) - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _shuffle1_dyn>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn shuffle1_dyn() { - let (null, non_null) = ptr_vals!($id<i32>); - - // alternating = [non_null, null, non_null, null, ...] - let mut alternating = $id::<i32>::splat(null); - for i in 0..$id::<i32>::lanes() { - if i % 2 == 0 { - alternating = alternating.replace(i, non_null); - } - } - - type Indices = <$id<i32> - as codegen::shuffle1_dyn::Shuffle1Dyn>::Indices; - // even = [0, 0, 2, 2, 4, 4, ..] - let even = { - let mut v = Indices::splat(0); - for i in 0..$id::<i32>::lanes() { - if i % 2 == 0 { - v = v.replace(i, (i as u8).into()); - } else { - v = v.replace(i, (i as u8 - 1).into()); - } - } - v - }; - // odd = [1, 1, 3, 3, 5, 5, ...] - let odd = { - let mut v = Indices::splat(0); - for i in 0..$id::<i32>::lanes() { - if i % 2 != 0 { - v = v.replace(i, (i as u8).into()); - } else { - v = v.replace(i, (i as u8 + 1).into()); - } - } - v - }; - - assert_eq!( - alternating.shuffle1_dyn(even), - $id::<i32>::splat(non_null) - ); - if $id::<i32>::lanes() > 1 { - assert_eq!( - alternating.shuffle1_dyn(odd), - $id::<i32>::splat(null) - ); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops.rs b/third_party/rust/packed_simd/src/api/ops.rs deleted file mode 100644 index f71c98795d..0000000000 --- a/third_party/rust/packed_simd/src/api/ops.rs +++ /dev/null @@ -1,32 +0,0 @@ -//! Implementation of the `ops` traits -#[macro_use] -mod vector_mask_bitwise; -#[macro_use] -mod scalar_mask_bitwise; - -#[macro_use] -mod vector_arithmetic; -#[macro_use] -mod scalar_arithmetic; - -#[macro_use] -mod vector_bitwise; -#[macro_use] -mod scalar_bitwise; - -#[macro_use] -mod vector_shifts; -#[macro_use] -mod scalar_shifts; - -#[macro_use] -mod vector_rotates; - -#[macro_use] -mod vector_neg; - -#[macro_use] -mod vector_int_min_max; - -#[macro_use] -mod vector_float_min_max; diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs b/third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs deleted file mode 100644 index da1a2037ea..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs +++ /dev/null @@ -1,203 +0,0 @@ -//! Vertical (lane-wise) vector-scalar / scalar-vector arithmetic operations. - -macro_rules! impl_ops_scalar_arithmetic { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::ops::Add<$elem_ty> for $id { - type Output = Self; - #[inline] - fn add(self, other: $elem_ty) -> Self { - self + $id::splat(other) - } - } - impl crate::ops::Add<$id> for $elem_ty { - type Output = $id; - #[inline] - fn add(self, other: $id) -> $id { - $id::splat(self) + other - } - } - - impl crate::ops::Sub<$elem_ty> for $id { - type Output = Self; - #[inline] - fn sub(self, other: $elem_ty) -> Self { - self - $id::splat(other) - } - } - impl crate::ops::Sub<$id> for $elem_ty { - type Output = $id; - #[inline] - fn sub(self, other: $id) -> $id { - $id::splat(self) - other - } - } - - impl crate::ops::Mul<$elem_ty> for $id { - type Output = Self; - #[inline] - fn mul(self, other: $elem_ty) -> Self { - self * $id::splat(other) - } - } - impl crate::ops::Mul<$id> for $elem_ty { - type Output = $id; - #[inline] - fn mul(self, other: $id) -> $id { - $id::splat(self) * other - } - } - - impl crate::ops::Div<$elem_ty> for $id { - type Output = Self; - #[inline] - fn div(self, other: $elem_ty) -> Self { - self / $id::splat(other) - } - } - impl crate::ops::Div<$id> for $elem_ty { - type Output = $id; - #[inline] - fn div(self, other: $id) -> $id { - $id::splat(self) / other - } - } - - impl crate::ops::Rem<$elem_ty> for $id { - type Output = Self; - #[inline] - fn rem(self, other: $elem_ty) -> Self { - self % $id::splat(other) - } - } - impl crate::ops::Rem<$id> for $elem_ty { - type Output = $id; - #[inline] - fn rem(self, other: $id) -> $id { - $id::splat(self) % other - } - } - - impl crate::ops::AddAssign<$elem_ty> for $id { - #[inline] - fn add_assign(&mut self, other: $elem_ty) { - *self = *self + other; - } - } - - impl crate::ops::SubAssign<$elem_ty> for $id { - #[inline] - fn sub_assign(&mut self, other: $elem_ty) { - *self = *self - other; - } - } - - impl crate::ops::MulAssign<$elem_ty> for $id { - #[inline] - fn mul_assign(&mut self, other: $elem_ty) { - *self = *self * other; - } - } - - impl crate::ops::DivAssign<$elem_ty> for $id { - #[inline] - fn div_assign(&mut self, other: $elem_ty) { - *self = *self / other; - } - } - - impl crate::ops::RemAssign<$elem_ty> for $id { - #[inline] - fn rem_assign(&mut self, other: $elem_ty) { - *self = *self % other; - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_scalar_arith>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ops_scalar_arithmetic() { - let zi = 0 as $elem_ty; - let oi = 1 as $elem_ty; - let ti = 2 as $elem_ty; - let fi = 4 as $elem_ty; - let z = $id::splat(zi); - let o = $id::splat(oi); - let t = $id::splat(ti); - let f = $id::splat(fi); - - // add - assert_eq!(zi + z, z); - assert_eq!(z + zi, z); - assert_eq!(oi + z, o); - assert_eq!(o + zi, o); - assert_eq!(ti + z, t); - assert_eq!(t + zi, t); - assert_eq!(ti + t, f); - assert_eq!(t + ti, f); - // sub - assert_eq!(zi - z, z); - assert_eq!(z - zi, z); - assert_eq!(oi - z, o); - assert_eq!(o - zi, o); - assert_eq!(ti - z, t); - assert_eq!(t - zi, t); - assert_eq!(fi - t, t); - assert_eq!(f - ti, t); - assert_eq!(f - o - o, t); - assert_eq!(f - oi - oi, t); - // mul - assert_eq!(zi * z, z); - assert_eq!(z * zi, z); - assert_eq!(zi * o, z); - assert_eq!(z * oi, z); - assert_eq!(zi * t, z); - assert_eq!(z * ti, z); - assert_eq!(oi * t, t); - assert_eq!(o * ti, t); - assert_eq!(ti * t, f); - assert_eq!(t * ti, f); - // div - assert_eq!(zi / o, z); - assert_eq!(z / oi, z); - assert_eq!(ti / o, t); - assert_eq!(t / oi, t); - assert_eq!(fi / o, f); - assert_eq!(f / oi, f); - assert_eq!(ti / t, o); - assert_eq!(t / ti, o); - assert_eq!(fi / t, t); - assert_eq!(f / ti, t); - // rem - assert_eq!(oi % o, z); - assert_eq!(o % oi, z); - assert_eq!(fi % t, z); - assert_eq!(f % ti, z); - - { - let mut v = z; - assert_eq!(v, z); - v += oi; // add_assign - assert_eq!(v, o); - v -= oi; // sub_assign - assert_eq!(v, z); - v = t; - v *= oi; // mul_assign - assert_eq!(v, t); - v *= ti; - assert_eq!(v, f); - v /= oi; // div_assign - assert_eq!(v, f); - v /= ti; - assert_eq!(v, t); - v %= ti; // rem_assign - assert_eq!(v, z); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs deleted file mode 100644 index 88216769ae..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs +++ /dev/null @@ -1,162 +0,0 @@ -//! Vertical (lane-wise) vector-scalar / scalar-vector bitwise operations. - -macro_rules! impl_ops_scalar_bitwise { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - impl crate::ops::BitXor<$elem_ty> for $id { - type Output = Self; - #[inline] - fn bitxor(self, other: $elem_ty) -> Self { - self ^ $id::splat(other) - } - } - impl crate::ops::BitXor<$id> for $elem_ty { - type Output = $id; - #[inline] - fn bitxor(self, other: $id) -> $id { - $id::splat(self) ^ other - } - } - - impl crate::ops::BitAnd<$elem_ty> for $id { - type Output = Self; - #[inline] - fn bitand(self, other: $elem_ty) -> Self { - self & $id::splat(other) - } - } - impl crate::ops::BitAnd<$id> for $elem_ty { - type Output = $id; - #[inline] - fn bitand(self, other: $id) -> $id { - $id::splat(self) & other - } - } - - impl crate::ops::BitOr<$elem_ty> for $id { - type Output = Self; - #[inline] - fn bitor(self, other: $elem_ty) -> Self { - self | $id::splat(other) - } - } - impl crate::ops::BitOr<$id> for $elem_ty { - type Output = $id; - #[inline] - fn bitor(self, other: $id) -> $id { - $id::splat(self) | other - } - } - - impl crate::ops::BitAndAssign<$elem_ty> for $id { - #[inline] - fn bitand_assign(&mut self, other: $elem_ty) { - *self = *self & other; - } - } - impl crate::ops::BitOrAssign<$elem_ty> for $id { - #[inline] - fn bitor_assign(&mut self, other: $elem_ty) { - *self = *self | other; - } - } - impl crate::ops::BitXorAssign<$elem_ty> for $id { - #[inline] - fn bitxor_assign(&mut self, other: $elem_ty) { - *self = *self ^ other; - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_scalar_bitwise>] { - use super::*; - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ops_scalar_bitwise() { - let zi = 0 as $elem_ty; - let oi = 1 as $elem_ty; - let ti = 2 as $elem_ty; - let z = $id::splat(zi); - let o = $id::splat(oi); - let t = $id::splat(ti); - - // BitAnd: - assert_eq!(oi & o, o); - assert_eq!(o & oi, o); - assert_eq!(oi & z, z); - assert_eq!(o & zi, z); - assert_eq!(zi & o, z); - assert_eq!(z & oi, z); - assert_eq!(zi & z, z); - assert_eq!(z & zi, z); - - assert_eq!(ti & t, t); - assert_eq!(t & ti, t); - assert_eq!(ti & o, z); - assert_eq!(t & oi, z); - assert_eq!(oi & t, z); - assert_eq!(o & ti, z); - - // BitOr: - assert_eq!(oi | o, o); - assert_eq!(o | oi, o); - assert_eq!(oi | z, o); - assert_eq!(o | zi, o); - assert_eq!(zi | o, o); - assert_eq!(z | oi, o); - assert_eq!(zi | z, z); - assert_eq!(z | zi, z); - - assert_eq!(ti | t, t); - assert_eq!(t | ti, t); - assert_eq!(zi | t, t); - assert_eq!(z | ti, t); - assert_eq!(ti | z, t); - assert_eq!(t | zi, t); - - // BitXOR: - assert_eq!(oi ^ o, z); - assert_eq!(o ^ oi, z); - assert_eq!(zi ^ z, z); - assert_eq!(z ^ zi, z); - assert_eq!(zi ^ o, o); - assert_eq!(z ^ oi, o); - assert_eq!(oi ^ z, o); - assert_eq!(o ^ zi, o); - - assert_eq!(ti ^ t, z); - assert_eq!(t ^ ti, z); - assert_eq!(ti ^ z, t); - assert_eq!(t ^ zi, t); - assert_eq!(zi ^ t, t); - assert_eq!(z ^ ti, t); - - { - // AndAssign: - let mut v = o; - v &= ti; - assert_eq!(v, z); - } - { - // OrAssign: - let mut v = z; - v |= oi; - assert_eq!(v, o); - } - { - // XORAssign: - let mut v = z; - v ^= oi; - assert_eq!(v, o); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs deleted file mode 100644 index 523a85207b..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs +++ /dev/null @@ -1,140 +0,0 @@ -//! Vertical (lane-wise) vector-vector bitwise operations. - -macro_rules! impl_ops_scalar_mask_bitwise { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - impl crate::ops::BitXor<bool> for $id { - type Output = Self; - #[inline] - fn bitxor(self, other: bool) -> Self { - self ^ $id::splat(other) - } - } - impl crate::ops::BitXor<$id> for bool { - type Output = $id; - #[inline] - fn bitxor(self, other: $id) -> $id { - $id::splat(self) ^ other - } - } - - impl crate::ops::BitAnd<bool> for $id { - type Output = Self; - #[inline] - fn bitand(self, other: bool) -> Self { - self & $id::splat(other) - } - } - impl crate::ops::BitAnd<$id> for bool { - type Output = $id; - #[inline] - fn bitand(self, other: $id) -> $id { - $id::splat(self) & other - } - } - - impl crate::ops::BitOr<bool> for $id { - type Output = Self; - #[inline] - fn bitor(self, other: bool) -> Self { - self | $id::splat(other) - } - } - impl crate::ops::BitOr<$id> for bool { - type Output = $id; - #[inline] - fn bitor(self, other: $id) -> $id { - $id::splat(self) | other - } - } - - impl crate::ops::BitAndAssign<bool> for $id { - #[inline] - fn bitand_assign(&mut self, other: bool) { - *self = *self & other; - } - } - impl crate::ops::BitOrAssign<bool> for $id { - #[inline] - fn bitor_assign(&mut self, other: bool) { - *self = *self | other; - } - } - impl crate::ops::BitXorAssign<bool> for $id { - #[inline] - fn bitxor_assign(&mut self, other: bool) { - *self = *self ^ other; - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_scalar_mask_bitwise>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ops_scalar_mask_bitwise() { - let ti = true; - let fi = false; - let t = $id::splat(ti); - let f = $id::splat(fi); - assert!(t != f); - assert!(!(t == f)); - - // BitAnd: - assert_eq!(ti & f, f); - assert_eq!(t & fi, f); - assert_eq!(fi & t, f); - assert_eq!(f & ti, f); - assert_eq!(ti & t, t); - assert_eq!(t & ti, t); - assert_eq!(fi & f, f); - assert_eq!(f & fi, f); - - // BitOr: - assert_eq!(ti | f, t); - assert_eq!(t | fi, t); - assert_eq!(fi | t, t); - assert_eq!(f | ti, t); - assert_eq!(ti | t, t); - assert_eq!(t | ti, t); - assert_eq!(fi | f, f); - assert_eq!(f | fi, f); - - // BitXOR: - assert_eq!(ti ^ f, t); - assert_eq!(t ^ fi, t); - assert_eq!(fi ^ t, t); - assert_eq!(f ^ ti, t); - assert_eq!(ti ^ t, f); - assert_eq!(t ^ ti, f); - assert_eq!(fi ^ f, f); - assert_eq!(f ^ fi, f); - - { - // AndAssign: - let mut v = f; - v &= ti; - assert_eq!(v, f); - } - { - // OrAssign: - let mut v = f; - v |= ti; - assert_eq!(v, t); - } - { - // XORAssign: - let mut v = f; - v ^= ti; - assert_eq!(v, t); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs b/third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs deleted file mode 100644 index 4a7a096263..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs +++ /dev/null @@ -1,106 +0,0 @@ -//! Vertical (lane-wise) vector-scalar shifts operations. - -macro_rules! impl_ops_scalar_shifts { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::ops::Shl<u32> for $id { - type Output = Self; - #[inline] - fn shl(self, other: u32) -> Self { - self << $id::splat(other as $elem_ty) - } - } - impl crate::ops::Shr<u32> for $id { - type Output = Self; - #[inline] - fn shr(self, other: u32) -> Self { - self >> $id::splat(other as $elem_ty) - } - } - - impl crate::ops::ShlAssign<u32> for $id { - #[inline] - fn shl_assign(&mut self, other: u32) { - *self = *self << other; - } - } - impl crate::ops::ShrAssign<u32> for $id { - #[inline] - fn shr_assign(&mut self, other: u32) { - *self = *self >> other; - } - } - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_scalar_shifts>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg_attr(any(target_arch = "s390x", target_arch = "sparc64"), - allow(unreachable_code, unused_variables) - )] - #[cfg(not(target_arch = "aarch64"))] - //~^ FIXME: https://github.com/rust-lang/packed_simd/issues/317 - fn ops_scalar_shifts() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - { - let zi = 0 as u32; - let oi = 1 as u32; - let ti = 2 as u32; - let maxi - = (mem::size_of::<$elem_ty>() * 8 - 1) as u32; - - // shr - assert_eq!(z >> zi, z); - assert_eq!(z >> oi, z); - assert_eq!(z >> ti, z); - assert_eq!(z >> ti, z); - - #[cfg(any(target_arch = "s390x", target_arch = "sparc64"))] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/13 - return; - } - - assert_eq!(o >> zi, o); - assert_eq!(t >> zi, t); - assert_eq!(f >> zi, f); - assert_eq!(f >> maxi, z); - - assert_eq!(o >> oi, z); - assert_eq!(t >> oi, o); - assert_eq!(t >> ti, z); - assert_eq!(f >> oi, t); - assert_eq!(f >> ti, o); - assert_eq!(f >> maxi, z); - - // shl - assert_eq!(z << zi, z); - assert_eq!(o << zi, o); - assert_eq!(t << zi, t); - assert_eq!(f << zi, f); - assert_eq!(f << maxi, z); - - assert_eq!(o << oi, t); - assert_eq!(o << ti, f); - assert_eq!(t << oi, f); - - { // shr_assign - let mut v = o; - v >>= oi; - assert_eq!(v, z); - } - { // shl_assign - let mut v = o; - v <<= oi; - assert_eq!(v, t); - } - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs b/third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs deleted file mode 100644 index 7057f52d03..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs +++ /dev/null @@ -1,148 +0,0 @@ -//! Vertical (lane-wise) vector-vector arithmetic operations. - -macro_rules! impl_ops_vector_arithmetic { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::ops::Add for $id { - type Output = Self; - #[inline] - fn add(self, other: Self) -> Self { - use crate::llvm::simd_add; - unsafe { Simd(simd_add(self.0, other.0)) } - } - } - - impl crate::ops::Sub for $id { - type Output = Self; - #[inline] - fn sub(self, other: Self) -> Self { - use crate::llvm::simd_sub; - unsafe { Simd(simd_sub(self.0, other.0)) } - } - } - - impl crate::ops::Mul for $id { - type Output = Self; - #[inline] - fn mul(self, other: Self) -> Self { - use crate::llvm::simd_mul; - unsafe { Simd(simd_mul(self.0, other.0)) } - } - } - - impl crate::ops::Div for $id { - type Output = Self; - #[inline] - fn div(self, other: Self) -> Self { - use crate::llvm::simd_div; - unsafe { Simd(simd_div(self.0, other.0)) } - } - } - - impl crate::ops::Rem for $id { - type Output = Self; - #[inline] - fn rem(self, other: Self) -> Self { - use crate::llvm::simd_rem; - unsafe { Simd(simd_rem(self.0, other.0)) } - } - } - - impl crate::ops::AddAssign for $id { - #[inline] - fn add_assign(&mut self, other: Self) { - *self = *self + other; - } - } - - impl crate::ops::SubAssign for $id { - #[inline] - fn sub_assign(&mut self, other: Self) { - *self = *self - other; - } - } - - impl crate::ops::MulAssign for $id { - #[inline] - fn mul_assign(&mut self, other: Self) { - *self = *self * other; - } - } - - impl crate::ops::DivAssign for $id { - #[inline] - fn div_assign(&mut self, other: Self) { - *self = *self / other; - } - } - - impl crate::ops::RemAssign for $id { - #[inline] - fn rem_assign(&mut self, other: Self) { - *self = *self % other; - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_vector_arith>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ops_vector_arithmetic() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - // add - assert_eq!(z + z, z); - assert_eq!(o + z, o); - assert_eq!(t + z, t); - assert_eq!(t + t, f); - // sub - assert_eq!(z - z, z); - assert_eq!(o - z, o); - assert_eq!(t - z, t); - assert_eq!(f - t, t); - assert_eq!(f - o - o, t); - // mul - assert_eq!(z * z, z); - assert_eq!(z * o, z); - assert_eq!(z * t, z); - assert_eq!(o * t, t); - assert_eq!(t * t, f); - // div - assert_eq!(z / o, z); - assert_eq!(t / o, t); - assert_eq!(f / o, f); - assert_eq!(t / t, o); - assert_eq!(f / t, t); - // rem - assert_eq!(o % o, z); - assert_eq!(f % t, z); - - { - let mut v = z; - assert_eq!(v, z); - v += o; // add_assign - assert_eq!(v, o); - v -= o; // sub_assign - assert_eq!(v, z); - v = t; - v *= o; // mul_assign - assert_eq!(v, t); - v *= t; - assert_eq!(v, f); - v /= o; // div_assign - assert_eq!(v, f); - v /= t; - assert_eq!(v, t); - v %= t; // rem_assign - assert_eq!(v, z); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs deleted file mode 100644 index 7be9603fa2..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs +++ /dev/null @@ -1,129 +0,0 @@ -//! Vertical (lane-wise) vector-vector bitwise operations. - -macro_rules! impl_ops_vector_bitwise { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - impl crate::ops::Not for $id { - type Output = Self; - #[inline] - fn not(self) -> Self { - Self::splat($true) ^ self - } - } - impl crate::ops::BitXor for $id { - type Output = Self; - #[inline] - fn bitxor(self, other: Self) -> Self { - use crate::llvm::simd_xor; - unsafe { Simd(simd_xor(self.0, other.0)) } - } - } - impl crate::ops::BitAnd for $id { - type Output = Self; - #[inline] - fn bitand(self, other: Self) -> Self { - use crate::llvm::simd_and; - unsafe { Simd(simd_and(self.0, other.0)) } - } - } - impl crate::ops::BitOr for $id { - type Output = Self; - #[inline] - fn bitor(self, other: Self) -> Self { - use crate::llvm::simd_or; - unsafe { Simd(simd_or(self.0, other.0)) } - } - } - impl crate::ops::BitAndAssign for $id { - #[inline] - fn bitand_assign(&mut self, other: Self) { - *self = *self & other; - } - } - impl crate::ops::BitOrAssign for $id { - #[inline] - fn bitor_assign(&mut self, other: Self) { - *self = *self | other; - } - } - impl crate::ops::BitXorAssign for $id { - #[inline] - fn bitxor_assign(&mut self, other: Self) { - *self = *self ^ other; - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_vector_bitwise>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ops_vector_bitwise() { - - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let m = $id::splat(!z.extract(0)); - - // Not: - assert_eq!(!z, m); - assert_eq!(!m, z); - - // BitAnd: - assert_eq!(o & o, o); - assert_eq!(o & z, z); - assert_eq!(z & o, z); - assert_eq!(z & z, z); - - assert_eq!(t & t, t); - assert_eq!(t & o, z); - assert_eq!(o & t, z); - - // BitOr: - assert_eq!(o | o, o); - assert_eq!(o | z, o); - assert_eq!(z | o, o); - assert_eq!(z | z, z); - - assert_eq!(t | t, t); - assert_eq!(z | t, t); - assert_eq!(t | z, t); - - // BitXOR: - assert_eq!(o ^ o, z); - assert_eq!(z ^ z, z); - assert_eq!(z ^ o, o); - assert_eq!(o ^ z, o); - - assert_eq!(t ^ t, z); - assert_eq!(t ^ z, t); - assert_eq!(z ^ t, t); - - { - // AndAssign: - let mut v = o; - v &= t; - assert_eq!(v, z); - } - { - // OrAssign: - let mut v = z; - v |= o; - assert_eq!(v, o); - } - { - // XORAssign: - let mut v = z; - v ^= o; - assert_eq!(v, o); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs b/third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs deleted file mode 100644 index 8310667b7a..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs +++ /dev/null @@ -1,74 +0,0 @@ -//! Vertical (lane-wise) vector `min` and `max` for floating-point vectors. - -macro_rules! impl_ops_vector_float_min_max { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Minimum of two vectors. - /// - /// Returns a new vector containing the minimum value of each of - /// the input vector lanes. - #[inline] - pub fn min(self, x: Self) -> Self { - use crate::llvm::simd_fmin; - unsafe { Simd(simd_fmin(self.0, x.0)) } - } - - /// Maximum of two vectors. - /// - /// Returns a new vector containing the maximum value of each of - /// the input vector lanes. - #[inline] - pub fn max(self, x: Self) -> Self { - use crate::llvm::simd_fmax; - unsafe { Simd(simd_fmax(self.0, x.0)) } - } - } - test_if!{ - $test_tt: - paste::item! { - #[cfg(not(any( - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/223 - all(target_arch = "mips", target_endian = "big"), - target_arch = "mips64", - )))] - pub mod [<$id _ops_vector_min_max>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn min_max() { - let n = crate::$elem_ty::NAN; - let o = $id::splat(1. as $elem_ty); - let t = $id::splat(2. as $elem_ty); - - let mut m = o; // [1., 2., 1., 2., ...] - let mut on = o; - for i in 0..$id::lanes() { - if i % 2 == 0 { - m = m.replace(i, 2. as $elem_ty); - on = on.replace(i, n); - } - } - - assert_eq!(o.min(t), o); - assert_eq!(t.min(o), o); - assert_eq!(m.min(o), o); - assert_eq!(o.min(m), o); - assert_eq!(m.min(t), m); - assert_eq!(t.min(m), m); - - assert_eq!(o.max(t), t); - assert_eq!(t.max(o), t); - assert_eq!(m.max(o), m); - assert_eq!(o.max(m), m); - assert_eq!(m.max(t), t); - assert_eq!(t.max(m), t); - - assert_eq!(on.min(o), o); - assert_eq!(o.min(on), o); - assert_eq!(on.max(o), o); - assert_eq!(o.max(on), o); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs b/third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs deleted file mode 100644 index 36ea98e6bf..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs +++ /dev/null @@ -1,57 +0,0 @@ -//! Vertical (lane-wise) vector `min` and `max` for integer vectors. - -macro_rules! impl_ops_vector_int_min_max { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Minimum of two vectors. - /// - /// Returns a new vector containing the minimum value of each of - /// the input vector lanes. - #[inline] - pub fn min(self, x: Self) -> Self { - self.lt(x).select(self, x) - } - - /// Maximum of two vectors. - /// - /// Returns a new vector containing the maximum value of each of - /// the input vector lanes. - #[inline] - pub fn max(self, x: Self) -> Self { - self.gt(x).select(self, x) - } - } - test_if!{$test_tt: - paste::item! { - pub mod [<$id _ops_vector_min_max>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn min_max() { - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - - let mut m = o; - for i in 0..$id::lanes() { - if i % 2 == 0 { - m = m.replace(i, 2 as $elem_ty); - } - } - assert_eq!(o.min(t), o); - assert_eq!(t.min(o), o); - assert_eq!(m.min(o), o); - assert_eq!(o.min(m), o); - assert_eq!(m.min(t), m); - assert_eq!(t.min(m), m); - - assert_eq!(o.max(t), t); - assert_eq!(t.max(o), t); - assert_eq!(m.max(o), m); - assert_eq!(o.max(m), m); - assert_eq!(m.max(t), t); - assert_eq!(t.max(m), t); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs deleted file mode 100644 index 295fc1ca81..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs +++ /dev/null @@ -1,116 +0,0 @@ -//! Vertical (lane-wise) vector-vector bitwise operations. - -macro_rules! impl_ops_vector_mask_bitwise { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $test_tt:tt | - ($true:expr, $false:expr) - ) => { - impl crate::ops::Not for $id { - type Output = Self; - #[inline] - fn not(self) -> Self { - Self::splat($true) ^ self - } - } - impl crate::ops::BitXor for $id { - type Output = Self; - #[inline] - fn bitxor(self, other: Self) -> Self { - use crate::llvm::simd_xor; - unsafe { Simd(simd_xor(self.0, other.0)) } - } - } - impl crate::ops::BitAnd for $id { - type Output = Self; - #[inline] - fn bitand(self, other: Self) -> Self { - use crate::llvm::simd_and; - unsafe { Simd(simd_and(self.0, other.0)) } - } - } - impl crate::ops::BitOr for $id { - type Output = Self; - #[inline] - fn bitor(self, other: Self) -> Self { - use crate::llvm::simd_or; - unsafe { Simd(simd_or(self.0, other.0)) } - } - } - impl crate::ops::BitAndAssign for $id { - #[inline] - fn bitand_assign(&mut self, other: Self) { - *self = *self & other; - } - } - impl crate::ops::BitOrAssign for $id { - #[inline] - fn bitor_assign(&mut self, other: Self) { - *self = *self | other; - } - } - impl crate::ops::BitXorAssign for $id { - #[inline] - fn bitxor_assign(&mut self, other: Self) { - *self = *self ^ other; - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_vector_mask_bitwise>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn ops_vector_mask_bitwise() { - let t = $id::splat(true); - let f = $id::splat(false); - assert!(t != f); - assert!(!(t == f)); - - // Not: - assert_eq!(!t, f); - assert_eq!(t, !f); - - // BitAnd: - assert_eq!(t & f, f); - assert_eq!(f & t, f); - assert_eq!(t & t, t); - assert_eq!(f & f, f); - - // BitOr: - assert_eq!(t | f, t); - assert_eq!(f | t, t); - assert_eq!(t | t, t); - assert_eq!(f | f, f); - - // BitXOR: - assert_eq!(t ^ f, t); - assert_eq!(f ^ t, t); - assert_eq!(t ^ t, f); - assert_eq!(f ^ f, f); - - { - // AndAssign: - let mut v = f; - v &= t; - assert_eq!(v, f); - } - { - // OrAssign: - let mut v = f; - v |= t; - assert_eq!(v, t); - } - { - // XORAssign: - let mut v = f; - v ^= t; - assert_eq!(v, t); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_neg.rs b/third_party/rust/packed_simd/src/api/ops/vector_neg.rs deleted file mode 100644 index e2d91fd2fe..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_neg.rs +++ /dev/null @@ -1,43 +0,0 @@ -//! Vertical (lane-wise) vector `Neg`. - -macro_rules! impl_ops_vector_neg { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::ops::Neg for $id { - type Output = Self; - #[inline] - fn neg(self) -> Self { - Self::splat(-1 as $elem_ty) * self - } - } - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_vector_neg>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn neg() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - let nz = $id::splat(-(0 as $elem_ty)); - let no = $id::splat(-(1 as $elem_ty)); - let nt = $id::splat(-(2 as $elem_ty)); - let nf = $id::splat(-(4 as $elem_ty)); - - assert_eq!(-z, nz); - assert_eq!(-o, no); - assert_eq!(-t, nt); - assert_eq!(-f, nf); - - assert_eq!(z, -nz); - assert_eq!(o, -no); - assert_eq!(t, -nt); - assert_eq!(f, -nf); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_rotates.rs b/third_party/rust/packed_simd/src/api/ops/vector_rotates.rs deleted file mode 100644 index 6c4bed72a2..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_rotates.rs +++ /dev/null @@ -1,92 +0,0 @@ -//! Vertical (lane-wise) vector rotates operations. -#![allow(unused)] - -macro_rules! impl_ops_vector_rotates { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Shifts the bits of each lane to the left by the specified - /// amount in the corresponding lane of `n`, wrapping the - /// truncated bits to the end of the resulting integer. - /// - /// Note: this is neither the same operation as `<<` nor equivalent - /// to `slice::rotate_left`. - #[inline] - pub fn rotate_left(self, n: $id) -> $id { - const LANE_WIDTH: $elem_ty = - crate::mem::size_of::<$elem_ty>() as $elem_ty * 8; - // Protect against undefined behavior for over-long bit shifts - let n = n % LANE_WIDTH; - (self << n) | (self >> ((LANE_WIDTH - n) % LANE_WIDTH)) - } - - /// Shifts the bits of each lane to the right by the specified - /// amount in the corresponding lane of `n`, wrapping the - /// truncated bits to the beginning of the resulting integer. - /// - /// Note: this is neither the same operation as `>>` nor equivalent - /// to `slice::rotate_right`. - #[inline] - pub fn rotate_right(self, n: $id) -> $id { - const LANE_WIDTH: $elem_ty = - crate::mem::size_of::<$elem_ty>() as $elem_ty * 8; - // Protect against undefined behavior for over-long bit shifts - let n = n % LANE_WIDTH; - (self >> n) | (self << ((LANE_WIDTH - n) % LANE_WIDTH)) - } - } - - test_if!{ - $test_tt: - paste::item! { - // FIXME: - // https://github.com/rust-lang-nursery/packed_simd/issues/75 - #[cfg(not(any( - target_arch = "s390x", - target_arch = "sparc64", - )))] - pub mod [<$id _ops_vector_rotate>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "aarch64"))] - //~^ FIXME: https://github.com/rust-lang/packed_simd/issues/317 - fn rotate_ops() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - let max = $id::splat( - (mem::size_of::<$elem_ty>() * 8 - 1) as $elem_ty); - - // rotate_right - assert_eq!(z.rotate_right(z), z); - assert_eq!(z.rotate_right(o), z); - assert_eq!(z.rotate_right(t), z); - - assert_eq!(o.rotate_right(z), o); - assert_eq!(t.rotate_right(z), t); - assert_eq!(f.rotate_right(z), f); - assert_eq!(f.rotate_right(max), f << 1); - - assert_eq!(o.rotate_right(o), o << max); - assert_eq!(t.rotate_right(o), o); - assert_eq!(t.rotate_right(t), o << max); - assert_eq!(f.rotate_right(o), t); - assert_eq!(f.rotate_right(t), o); - - // rotate_left - assert_eq!(z.rotate_left(z), z); - assert_eq!(o.rotate_left(z), o); - assert_eq!(t.rotate_left(z), t); - assert_eq!(f.rotate_left(z), f); - assert_eq!(f.rotate_left(max), t); - - assert_eq!(o.rotate_left(o), t); - assert_eq!(o.rotate_left(t), f); - assert_eq!(t.rotate_left(o), f); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_shifts.rs b/third_party/rust/packed_simd/src/api/ops/vector_shifts.rs deleted file mode 100644 index 8bb5ac2fc0..0000000000 --- a/third_party/rust/packed_simd/src/api/ops/vector_shifts.rs +++ /dev/null @@ -1,106 +0,0 @@ -//! Vertical (lane-wise) vector-vector shifts operations. - -macro_rules! impl_ops_vector_shifts { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl crate::ops::Shl<$id> for $id { - type Output = Self; - #[inline] - fn shl(self, other: Self) -> Self { - use crate::llvm::simd_shl; - unsafe { Simd(simd_shl(self.0, other.0)) } - } - } - impl crate::ops::Shr<$id> for $id { - type Output = Self; - #[inline] - fn shr(self, other: Self) -> Self { - use crate::llvm::simd_shr; - unsafe { Simd(simd_shr(self.0, other.0)) } - } - } - impl crate::ops::ShlAssign<$id> for $id { - #[inline] - fn shl_assign(&mut self, other: Self) { - *self = *self << other; - } - } - impl crate::ops::ShrAssign<$id> for $id { - #[inline] - fn shr_assign(&mut self, other: Self) { - *self = *self >> other; - } - } - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _ops_vector_shifts>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg_attr(any(target_arch = "s390x", target_arch = "sparc64"), - allow(unreachable_code, unused_variables) - )] - #[cfg(not(target_arch = "aarch64"))] - //~^ FIXME: https://github.com/rust-lang/packed_simd/issues/317 - fn ops_vector_shifts() { - let z = $id::splat(0 as $elem_ty); - let o = $id::splat(1 as $elem_ty); - let t = $id::splat(2 as $elem_ty); - let f = $id::splat(4 as $elem_ty); - - let max =$id::splat( - (mem::size_of::<$elem_ty>() * 8 - 1) as $elem_ty - ); - - // shr - assert_eq!(z >> z, z); - assert_eq!(z >> o, z); - assert_eq!(z >> t, z); - assert_eq!(z >> t, z); - - #[cfg(any(target_arch = "s390x", target_arch = "sparc64"))] { - // FIXME: rust produces bad codegen for shifts: - // https://github.com/rust-lang-nursery/packed_simd/issues/13 - return; - } - - assert_eq!(o >> z, o); - assert_eq!(t >> z, t); - assert_eq!(f >> z, f); - assert_eq!(f >> max, z); - - assert_eq!(o >> o, z); - assert_eq!(t >> o, o); - assert_eq!(t >> t, z); - assert_eq!(f >> o, t); - assert_eq!(f >> t, o); - assert_eq!(f >> max, z); - - // shl - assert_eq!(z << z, z); - assert_eq!(o << z, o); - assert_eq!(t << z, t); - assert_eq!(f << z, f); - assert_eq!(f << max, z); - - assert_eq!(o << o, t); - assert_eq!(o << t, f); - assert_eq!(t << o, f); - - { - // shr_assign - let mut v = o; - v >>= o; - assert_eq!(v, z); - } - { - // shl_assign - let mut v = o; - v <<= o; - assert_eq!(v, t); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/ptr.rs b/third_party/rust/packed_simd/src/api/ptr.rs deleted file mode 100644 index d2e523a49f..0000000000 --- a/third_party/rust/packed_simd/src/api/ptr.rs +++ /dev/null @@ -1,4 +0,0 @@ -//! Vector of pointers - -#[macro_use] -mod gather_scatter; diff --git a/third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs b/third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs deleted file mode 100644 index 374482ac31..0000000000 --- a/third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs +++ /dev/null @@ -1,216 +0,0 @@ -//! Implements masked gather and scatters for vectors of pointers - -macro_rules! impl_ptr_read { - ([$elem_ty:ty; $elem_count:expr]: $id:ident, $mask_ty:ident - | $test_tt:tt) => { - impl<T> $id<T> - where - [T; $elem_count]: sealed::SimdArray, - { - /// Reads selected vector elements from memory. - /// - /// Instantiates a new vector by reading the values from `self` for - /// those lanes whose `mask` is `true`, and using the elements of - /// `value` otherwise. - /// - /// No memory is accessed for those lanes of `self` whose `mask` is - /// `false`. - /// - /// # Safety - /// - /// This method is unsafe because it dereferences raw pointers. The - /// pointers must be aligned to `mem::align_of::<T>()`. - #[inline] - pub unsafe fn read<M>( - self, - mask: Simd<[M; $elem_count]>, - value: Simd<[T; $elem_count]>, - ) -> Simd<[T; $elem_count]> - where - M: sealed::Mask, - [M; $elem_count]: sealed::SimdArray, - { - use crate::llvm::simd_gather; - Simd(simd_gather(value.0, self.0, mask.0)) - } - } - - test_if! { - $test_tt: - paste::item! { - mod [<$id _read>] { - use super::*; - #[test] - fn read() { - let mut v = [0_i32; $elem_count]; - for i in 0..$elem_count { - v[i] = i as i32; - } - - let mut ptr = $id::<i32>::null(); - - for i in 0..$elem_count { - ptr = ptr.replace(i, - &v[i] as *const i32 as *mut i32 - ); - } - - // all mask elements are true: - let mask = $mask_ty::splat(true); - let def = Simd::<[i32; $elem_count]>::splat(42_i32); - let r: Simd<[i32; $elem_count]> = unsafe { - ptr.read(mask, def) - }; - assert_eq!( - r, - Simd::<[i32; $elem_count]>::from_slice_unaligned( - &v - ) - ); - - let mut mask = mask; - for i in 0..$elem_count { - if i % 2 != 0 { - mask = mask.replace(i, false); - } - } - - // even mask elements are true, odd ones are false: - let r: Simd<[i32; $elem_count]> = unsafe { - ptr.read(mask, def) - }; - let mut e = v; - for i in 0..$elem_count { - if i % 2 != 0 { - e[i] = 42; - } - } - assert_eq!( - r, - Simd::<[i32; $elem_count]>::from_slice_unaligned( - &e - ) - ); - - // all mask elements are false: - let mask = $mask_ty::splat(false); - let def = Simd::<[i32; $elem_count]>::splat(42_i32); - let r: Simd<[i32; $elem_count]> = unsafe { - ptr.read(mask, def) } - ; - assert_eq!(r, def); - } - } - } - } - }; -} - -macro_rules! impl_ptr_write { - ([$elem_ty:ty; $elem_count:expr]: $id:ident, $mask_ty:ident - | $test_tt:tt) => { - impl<T> $id<T> - where - [T; $elem_count]: sealed::SimdArray, - { - /// Writes selected vector elements to memory. - /// - /// Writes the lanes of `values` for which the mask is `true` to - /// their corresponding memory addresses in `self`. - /// - /// No memory is accessed for those lanes of `self` whose `mask` is - /// `false`. - /// - /// Overlapping memory addresses of `self` are written to in order - /// from the lest-significant to the most-significant element. - /// - /// # Safety - /// - /// This method is unsafe because it dereferences raw pointers. The - /// pointers must be aligned to `mem::align_of::<T>()`. - #[inline] - pub unsafe fn write<M>(self, mask: Simd<[M; $elem_count]>, value: Simd<[T; $elem_count]>) - where - M: sealed::Mask, - [M; $elem_count]: sealed::SimdArray, - { - use crate::llvm::simd_scatter; - simd_scatter(value.0, self.0, mask.0) - } - } - - test_if! { - $test_tt: - paste::item! { - mod [<$id _write>] { - use super::*; - #[test] - fn write() { - // forty_two = [42, 42, 42, ...] - let forty_two - = Simd::<[i32; $elem_count]>::splat(42_i32); - - // This test will write to this array - let mut arr = [0_i32; $elem_count]; - for i in 0..$elem_count { - arr[i] = i as i32; - } - // arr = [0, 1, 2, ...] - - let mut ptr = $id::<i32>::null(); - for i in 0..$elem_count { - ptr = ptr.replace(i, unsafe { - arr.as_ptr().add(i) as *mut i32 - }); - } - // ptr = [&arr[0], &arr[1], ...] - - // write `forty_two` to all elements of `v` - { - let backup = arr; - unsafe { - ptr.write($mask_ty::splat(true), forty_two) - }; - assert_eq!(arr, [42_i32; $elem_count]); - arr = backup; // arr = [0, 1, 2, ...] - } - - // write 42 to even elements of arr: - { - // set odd elements of the mask to false - let mut mask = $mask_ty::splat(true); - for i in 0..$elem_count { - if i % 2 != 0 { - mask = mask.replace(i, false); - } - } - // mask = [true, false, true, false, ...] - - // expected result r = [42, 1, 42, 3, 42, 5, ...] - let mut r = arr; - for i in 0..$elem_count { - if i % 2 == 0 { - r[i] = 42; - } - } - - let backup = arr; - unsafe { ptr.write(mask, forty_two) }; - assert_eq!(arr, r); - arr = backup; // arr = [0, 1, 2, 3, ...] - } - - // write 42 to no elements of arr - { - let backup = arr; - unsafe { - ptr.write($mask_ty::splat(false), forty_two) - }; - assert_eq!(arr, backup); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/reductions.rs b/third_party/rust/packed_simd/src/api/reductions.rs deleted file mode 100644 index 54d2f0cc7f..0000000000 --- a/third_party/rust/packed_simd/src/api/reductions.rs +++ /dev/null @@ -1,12 +0,0 @@ -//! Reductions - -#[macro_use] -mod float_arithmetic; -#[macro_use] -mod integer_arithmetic; -#[macro_use] -mod bitwise; -#[macro_use] -mod mask; -#[macro_use] -mod min_max; diff --git a/third_party/rust/packed_simd/src/api/reductions/bitwise.rs b/third_party/rust/packed_simd/src/api/reductions/bitwise.rs deleted file mode 100644 index 5bad4f474b..0000000000 --- a/third_party/rust/packed_simd/src/api/reductions/bitwise.rs +++ /dev/null @@ -1,151 +0,0 @@ -//! Implements portable horizontal bitwise vector reductions. -#![allow(unused)] - -macro_rules! impl_reduction_bitwise { - ( - [$elem_ty:ident; $elem_count:expr]: - $id:ident | $ielem_ty:ident | $test_tt:tt | - ($convert:expr) | - ($true:expr, $false:expr) - ) => { - impl $id { - /// Lane-wise bitwise `and` of the vector elements. - /// - /// Note: if the vector has one lane, the first element of the - /// vector is returned. - #[inline] - pub fn and(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_and; - let r: $ielem_ty = unsafe { simd_reduce_and(self.0) }; - $convert(r) - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on aarch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x &= self.extract(i) as $elem_ty; - } - x - } - } - - /// Lane-wise bitwise `or` of the vector elements. - /// - /// Note: if the vector has one lane, the first element of the - /// vector is returned. - #[inline] - pub fn or(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_or; - let r: $ielem_ty = unsafe { simd_reduce_or(self.0) }; - $convert(r) - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on aarch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x |= self.extract(i) as $elem_ty; - } - x - } - } - - /// Lane-wise bitwise `xor` of the vector elements. - /// - /// Note: if the vector has one lane, the first element of the - /// vector is returned. - #[inline] - pub fn xor(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_xor; - let r: $ielem_ty = unsafe { simd_reduce_xor(self.0) }; - $convert(r) - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on aarch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x ^= self.extract(i) as $elem_ty; - } - x - } - } - } - - test_if!{ - $test_tt: - paste::item! { - pub mod [<$id _reduction_bitwise>] { - use super::*; - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn and() { - let v = $id::splat($false); - assert_eq!(v.and(), $false); - let v = $id::splat($true); - assert_eq!(v.and(), $true); - let v = $id::splat($false); - let v = v.replace(0, $true); - if $id::lanes() > 1 { - assert_eq!(v.and(), $false); - } else { - assert_eq!(v.and(), $true); - } - let v = $id::splat($true); - let v = v.replace(0, $false); - assert_eq!(v.and(), $false); - - } - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn or() { - let v = $id::splat($false); - assert_eq!(v.or(), $false); - let v = $id::splat($true); - assert_eq!(v.or(), $true); - let v = $id::splat($false); - let v = v.replace(0, $true); - assert_eq!(v.or(), $true); - let v = $id::splat($true); - let v = v.replace(0, $false); - if $id::lanes() > 1 { - assert_eq!(v.or(), $true); - } else { - assert_eq!(v.or(), $false); - } - } - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn xor() { - let v = $id::splat($false); - assert_eq!(v.xor(), $false); - let v = $id::splat($true); - if $id::lanes() > 1 { - assert_eq!(v.xor(), $false); - } else { - assert_eq!(v.xor(), $true); - } - let v = $id::splat($false); - let v = v.replace(0, $true); - assert_eq!(v.xor(), $true); - let v = $id::splat($true); - let v = v.replace(0, $false); - if $id::lanes() > 1 { - assert_eq!(v.xor(), $true); - } else { - assert_eq!(v.xor(), $false); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs b/third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs deleted file mode 100644 index 9dc8783dbb..0000000000 --- a/third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs +++ /dev/null @@ -1,313 +0,0 @@ -//! Implements portable horizontal float vector arithmetic reductions. - -macro_rules! impl_reduction_float_arithmetic { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Horizontal sum of the vector elements. - /// - /// The intrinsic performs a tree-reduction of the vector elements. - /// That is, for an 8 element vector: - /// - /// > ((x0 + x1) + (x2 + x3)) + ((x4 + x5) + (x6 + x7)) - /// - /// If one of the vector element is `NaN` the reduction returns - /// `NaN`. The resulting `NaN` is not required to be equal to any - /// of the `NaN`s in the vector. - #[inline] - pub fn sum(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_add_ordered; - unsafe { simd_reduce_add_ordered(self.0, 0 as $elem_ty) } - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on AArch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x += self.extract(i) as $elem_ty; - } - x - } - } - - /// Horizontal product of the vector elements. - /// - /// The intrinsic performs a tree-reduction of the vector elements. - /// That is, for an 8 element vector: - /// - /// > ((x0 * x1) * (x2 * x3)) * ((x4 * x5) * (x6 * x7)) - /// - /// If one of the vector element is `NaN` the reduction returns - /// `NaN`. The resulting `NaN` is not required to be equal to any - /// of the `NaN`s in the vector. - #[inline] - pub fn product(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_mul_ordered; - unsafe { simd_reduce_mul_ordered(self.0, 1 as $elem_ty) } - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on AArch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x *= self.extract(i) as $elem_ty; - } - x - } - } - } - - impl crate::iter::Sum for $id { - #[inline] - fn sum<I: Iterator<Item = $id>>(iter: I) -> $id { - iter.fold($id::splat(0.), crate::ops::Add::add) - } - } - - impl crate::iter::Product for $id { - #[inline] - fn product<I: Iterator<Item = $id>>(iter: I) -> $id { - iter.fold($id::splat(1.), crate::ops::Mul::mul) - } - } - - impl<'a> crate::iter::Sum<&'a $id> for $id { - #[inline] - fn sum<I: Iterator<Item = &'a $id>>(iter: I) -> $id { - iter.fold($id::splat(0.), |a, b| crate::ops::Add::add(a, *b)) - } - } - - impl<'a> crate::iter::Product<&'a $id> for $id { - #[inline] - fn product<I: Iterator<Item = &'a $id>>(iter: I) -> $id { - iter.fold($id::splat(1.), |a, b| crate::ops::Mul::mul(a, *b)) - } - } - - test_if! { - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _reduction_float_arith>] { - use super::*; - fn alternating(x: usize) -> $id { - let mut v = $id::splat(1 as $elem_ty); - for i in 0..$id::lanes() { - if i % x == 0 { - v = v.replace(i, 2 as $elem_ty); - } - } - v - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn sum() { - let v = $id::splat(0 as $elem_ty); - assert_eq!(v.sum(), 0 as $elem_ty); - let v = $id::splat(1 as $elem_ty); - assert_eq!(v.sum(), $id::lanes() as $elem_ty); - let v = alternating(2); - assert_eq!( - v.sum(), - ($id::lanes() / 2 + $id::lanes()) as $elem_ty - ); - } - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn product() { - let v = $id::splat(0 as $elem_ty); - assert_eq!(v.product(), 0 as $elem_ty); - let v = $id::splat(1 as $elem_ty); - assert_eq!(v.product(), 1 as $elem_ty); - let f = match $id::lanes() { - 64 => 16, - 32 => 8, - 16 => 4, - _ => 2, - }; - let v = alternating(f); - assert_eq!( - v.product(), - (2_usize.pow(($id::lanes() / f) as u32) - as $elem_ty) - ); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[allow(unreachable_code)] - fn sum_nan() { - // FIXME: https://bugs.llvm.org/show_bug.cgi?id=36732 - // https://github.com/rust-lang-nursery/packed_simd/issues/6 - return; - - let n0 = crate::$elem_ty::NAN; - let v0 = $id::splat(-3.0); - for i in 0..$id::lanes() { - let mut v = v0.replace(i, n0); - // If the vector contains a NaN the result is NaN: - assert!( - v.sum().is_nan(), - "nan at {} => {} | {:?}", - i, - v.sum(), - v - ); - for j in 0..i { - v = v.replace(j, n0); - assert!(v.sum().is_nan()); - } - } - let v = $id::splat(n0); - assert!(v.sum().is_nan(), "all nans | {:?}", v); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[allow(unreachable_code)] - fn product_nan() { - // FIXME: https://bugs.llvm.org/show_bug.cgi?id=36732 - // https://github.com/rust-lang-nursery/packed_simd/issues/6 - return; - - let n0 = crate::$elem_ty::NAN; - let v0 = $id::splat(-3.0); - for i in 0..$id::lanes() { - let mut v = v0.replace(i, n0); - // If the vector contains a NaN the result is NaN: - assert!( - v.product().is_nan(), - "nan at {} => {} | {:?}", - i, - v.product(), - v - ); - for j in 0..i { - v = v.replace(j, n0); - assert!(v.product().is_nan()); - } - } - let v = $id::splat(n0); - assert!(v.product().is_nan(), "all nans | {:?}", v); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[allow(unused, dead_code)] - fn sum_roundoff() { - // Performs a tree-reduction - fn tree_reduce_sum(a: &[$elem_ty]) -> $elem_ty { - assert!(!a.is_empty()); - if a.len() == 1 { - a[0] - } else if a.len() == 2 { - a[0] + a[1] - } else { - let mid = a.len() / 2; - let (left, right) = a.split_at(mid); - tree_reduce_sum(left) + tree_reduce_sum(right) - } - } - - let mut start = crate::$elem_ty::EPSILON; - let mut scalar_reduction = 0. as $elem_ty; - - let mut v = $id::splat(0. as $elem_ty); - for i in 0..$id::lanes() { - let c = if i % 2 == 0 { 1e3 } else { -1. }; - start *= ::core::$elem_ty::consts::PI * c; - scalar_reduction += start; - v = v.replace(i, start); - } - let simd_reduction = v.sum(); - - let mut a = [0. as $elem_ty; $id::lanes()]; - v.write_to_slice_unaligned(&mut a); - let tree_reduction = tree_reduce_sum(&a); - - // tolerate 1 ULP difference: - let red_bits = simd_reduction.to_bits(); - let tree_bits = tree_reduction.to_bits(); - assert!( - if red_bits > tree_bits { - red_bits - tree_bits - } else { - tree_bits - red_bits - } < 2, - "vector: {:?} | simd_reduction: {:?} | \ -tree_reduction: {} | scalar_reduction: {}", - v, - simd_reduction, - tree_reduction, - scalar_reduction - ); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[allow(unused, dead_code)] - fn product_roundoff() { - use ::core::convert::TryInto; - // Performs a tree-reduction - fn tree_reduce_product(a: &[$elem_ty]) -> $elem_ty { - assert!(!a.is_empty()); - if a.len() == 1 { - a[0] - } else if a.len() == 2 { - a[0] * a[1] - } else { - let mid = a.len() / 2; - let (left, right) = a.split_at(mid); - tree_reduce_product(left) - * tree_reduce_product(right) - } - } - - let mut start = crate::$elem_ty::EPSILON; - let mut scalar_reduction = 1. as $elem_ty; - - let mut v = $id::splat(0. as $elem_ty); - for i in 0..$id::lanes() { - let c = if i % 2 == 0 { 1e3 } else { -1. }; - start *= ::core::$elem_ty::consts::PI * c; - scalar_reduction *= start; - v = v.replace(i, start); - } - let simd_reduction = v.product(); - - let mut a = [0. as $elem_ty; $id::lanes()]; - v.write_to_slice_unaligned(&mut a); - let tree_reduction = tree_reduce_product(&a); - - // FIXME: Too imprecise, even only for product(f32x8). - // Figure out how to narrow this down. - let ulp_limit = $id::lanes() / 2; - let red_bits = simd_reduction.to_bits(); - let tree_bits = tree_reduction.to_bits(); - assert!( - if red_bits > tree_bits { - red_bits - tree_bits - } else { - tree_bits - red_bits - } < ulp_limit.try_into().unwrap(), - "vector: {:?} | simd_reduction: {:?} | \ -tree_reduction: {} | scalar_reduction: {}", - v, - simd_reduction, - tree_reduction, - scalar_reduction - ); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs b/third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs deleted file mode 100644 index e99e6cb5d7..0000000000 --- a/third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs +++ /dev/null @@ -1,193 +0,0 @@ -//! Implements portable horizontal integer vector arithmetic reductions. - -macro_rules! impl_reduction_integer_arithmetic { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $ielem_ty:ident - | $test_tt:tt) => { - impl $id { - /// Horizontal wrapping sum of the vector elements. - /// - /// The intrinsic performs a tree-reduction of the vector elements. - /// That is, for an 8 element vector: - /// - /// > ((x0 + x1) + (x2 + x3)) + ((x4 + x5) + (x6 + x7)) - /// - /// If an operation overflows it returns the mathematical result - /// modulo `2^n` where `n` is the number of times it overflows. - #[inline] - pub fn wrapping_sum(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_add_ordered; - let v: $ielem_ty = unsafe { simd_reduce_add_ordered(self.0, 0 as $ielem_ty) }; - v as $elem_ty - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on AArch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x = x.wrapping_add(self.extract(i) as $elem_ty); - } - x - } - } - - /// Horizontal wrapping product of the vector elements. - /// - /// The intrinsic performs a tree-reduction of the vector elements. - /// That is, for an 8 element vector: - /// - /// > ((x0 * x1) * (x2 * x3)) * ((x4 * x5) * (x6 * x7)) - /// - /// If an operation overflows it returns the mathematical result - /// modulo `2^n` where `n` is the number of times it overflows. - #[inline] - pub fn wrapping_product(self) -> $elem_ty { - #[cfg(not(target_arch = "aarch64"))] - { - use crate::llvm::simd_reduce_mul_ordered; - let v: $ielem_ty = unsafe { simd_reduce_mul_ordered(self.0, 1 as $ielem_ty) }; - v as $elem_ty - } - #[cfg(target_arch = "aarch64")] - { - // FIXME: broken on AArch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - let mut x = self.extract(0) as $elem_ty; - for i in 1..$id::lanes() { - x = x.wrapping_mul(self.extract(i) as $elem_ty); - } - x - } - } - } - - impl crate::iter::Sum for $id { - #[inline] - fn sum<I: Iterator<Item = $id>>(iter: I) -> $id { - iter.fold($id::splat(0), crate::ops::Add::add) - } - } - - impl crate::iter::Product for $id { - #[inline] - fn product<I: Iterator<Item = $id>>(iter: I) -> $id { - iter.fold($id::splat(1), crate::ops::Mul::mul) - } - } - - impl<'a> crate::iter::Sum<&'a $id> for $id { - #[inline] - fn sum<I: Iterator<Item = &'a $id>>(iter: I) -> $id { - iter.fold($id::splat(0), |a, b| crate::ops::Add::add(a, *b)) - } - } - - impl<'a> crate::iter::Product<&'a $id> for $id { - #[inline] - fn product<I: Iterator<Item = &'a $id>>(iter: I) -> $id { - iter.fold($id::splat(1), |a, b| crate::ops::Mul::mul(a, *b)) - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _reduction_int_arith>] { - use super::*; - - fn alternating(x: usize) -> $id { - let mut v = $id::splat(1 as $elem_ty); - for i in 0..$id::lanes() { - if i % x == 0 { - v = v.replace(i, 2 as $elem_ty); - } - } - v - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn wrapping_sum() { - let v = $id::splat(0 as $elem_ty); - assert_eq!(v.wrapping_sum(), 0 as $elem_ty); - let v = $id::splat(1 as $elem_ty); - assert_eq!(v.wrapping_sum(), $id::lanes() as $elem_ty); - let v = alternating(2); - if $id::lanes() > 1 { - assert_eq!( - v.wrapping_sum(), - ($id::lanes() / 2 + $id::lanes()) as $elem_ty - ); - } else { - assert_eq!( - v.wrapping_sum(), - 2 as $elem_ty - ); - } - } - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn wrapping_sum_overflow() { - let start = $elem_ty::max_value() - - ($id::lanes() as $elem_ty / 2); - - let v = $id::splat(start as $elem_ty); - let vwrapping_sum = v.wrapping_sum(); - - let mut wrapping_sum = start; - for _ in 1..$id::lanes() { - wrapping_sum = wrapping_sum.wrapping_add(start); - } - assert_eq!(wrapping_sum, vwrapping_sum, "v = {:?}", v); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn wrapping_product() { - let v = $id::splat(0 as $elem_ty); - assert_eq!(v.wrapping_product(), 0 as $elem_ty); - let v = $id::splat(1 as $elem_ty); - assert_eq!(v.wrapping_product(), 1 as $elem_ty); - let f = match $id::lanes() { - 64 => 16, - 32 => 8, - 16 => 4, - _ => 2, - }; - let v = alternating(f); - if $id::lanes() > 1 { - assert_eq!( - v.wrapping_product(), - (2_usize.pow(($id::lanes() / f) as u32) - as $elem_ty) - ); - } else { - assert_eq!( - v.wrapping_product(), - 2 as $elem_ty - ); - } - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn wrapping_product_overflow() { - let start = $elem_ty::max_value() - - ($id::lanes() as $elem_ty / 2); - - let v = $id::splat(start as $elem_ty); - let vmul = v.wrapping_product(); - - let mut mul = start; - for _ in 1..$id::lanes() { - mul = mul.wrapping_mul(start); - } - assert_eq!(mul, vmul, "v = {:?}", v); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/reductions/mask.rs b/third_party/rust/packed_simd/src/api/reductions/mask.rs deleted file mode 100644 index 0dd6a84e7e..0000000000 --- a/third_party/rust/packed_simd/src/api/reductions/mask.rs +++ /dev/null @@ -1,89 +0,0 @@ -//! Implements portable horizontal mask reductions. - -macro_rules! impl_reduction_mask { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Are `all` vector lanes `true`? - #[inline] - pub fn all(self) -> bool { - unsafe { crate::codegen::reductions::mask::All::all(self) } - } - /// Is `any` vector lane `true`? - #[inline] - pub fn any(self) -> bool { - unsafe { crate::codegen::reductions::mask::Any::any(self) } - } - /// Are `all` vector lanes `false`? - #[inline] - pub fn none(self) -> bool { - !self.any() - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _reduction>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn all() { - let a = $id::splat(true); - assert!(a.all()); - let a = $id::splat(false); - assert!(!a.all()); - - if $id::lanes() > 1 { - for i in 0..$id::lanes() { - let mut a = $id::splat(true); - a = a.replace(i, false); - assert!(!a.all()); - let mut a = $id::splat(false); - a = a.replace(i, true); - assert!(!a.all()); - } - } - } - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn any() { - let a = $id::splat(true); - assert!(a.any()); - let a = $id::splat(false); - assert!(!a.any()); - - if $id::lanes() > 1 { - for i in 0..$id::lanes() { - let mut a = $id::splat(true); - a = a.replace(i, false); - assert!(a.any()); - let mut a = $id::splat(false); - a = a.replace(i, true); - assert!(a.any()); - } - } - } - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn none() { - let a = $id::splat(true); - assert!(!a.none()); - let a = $id::splat(false); - assert!(a.none()); - - if $id::lanes() > 1 { - for i in 0..$id::lanes() { - let mut a = $id::splat(true); - a = a.replace(i, false); - assert!(!a.none()); - let mut a = $id::splat(false); - a = a.replace(i, true); - assert!(!a.none()); - } - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/reductions/min_max.rs b/third_party/rust/packed_simd/src/api/reductions/min_max.rs deleted file mode 100644 index a3ce13a451..0000000000 --- a/third_party/rust/packed_simd/src/api/reductions/min_max.rs +++ /dev/null @@ -1,360 +0,0 @@ -//! Implements portable horizontal vector min/max reductions. - -macro_rules! impl_reduction_min_max { - ([$elem_ty:ident; $elem_count:expr]: $id:ident - | $ielem_ty:ident | $test_tt:tt) => { - impl $id { - /// Largest vector element value. - #[inline] - pub fn max_element(self) -> $elem_ty { - #[cfg(not(any( - target_arch = "aarch64", - target_arch = "arm", - target_arch = "powerpc64", - target_arch = "wasm32", - )))] - { - use crate::llvm::simd_reduce_max; - let v: $ielem_ty = unsafe { simd_reduce_max(self.0) }; - v as $elem_ty - } - #[cfg(any( - target_arch = "aarch64", - target_arch = "arm", - target_arch = "powerpc64", - target_arch = "wasm32", - ))] - { - // FIXME: broken on AArch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - // FIXME: broken on WASM32 - // https://github.com/rust-lang-nursery/packed_simd/issues/91 - let mut x = self.extract(0); - for i in 1..$id::lanes() { - x = x.max(self.extract(i)); - } - x - } - } - - /// Smallest vector element value. - #[inline] - pub fn min_element(self) -> $elem_ty { - #[cfg(not(any( - target_arch = "aarch64", - target_arch = "arm", - all(target_arch = "x86", not(target_feature = "sse2")), - target_arch = "powerpc64", - target_arch = "wasm32", - ),))] - { - use crate::llvm::simd_reduce_min; - let v: $ielem_ty = unsafe { simd_reduce_min(self.0) }; - v as $elem_ty - } - #[cfg(any( - target_arch = "aarch64", - target_arch = "arm", - all(target_arch = "x86", not(target_feature = "sse2")), - target_arch = "powerpc64", - target_arch = "wasm32", - ))] - { - // FIXME: broken on AArch64 - // https://github.com/rust-lang-nursery/packed_simd/issues/15 - // FIXME: broken on i586-unknown-linux-gnu - // https://github.com/rust-lang-nursery/packed_simd/issues/22 - // FIXME: broken on WASM32 - // https://github.com/rust-lang-nursery/packed_simd/issues/91 - let mut x = self.extract(0); - for i in 1..$id::lanes() { - x = x.min(self.extract(i)); - } - x - } - } - } - test_if! {$test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _reduction_min_max>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - pub fn max_element() { - let v = $id::splat(0 as $elem_ty); - assert_eq!(v.max_element(), 0 as $elem_ty); - if $id::lanes() > 1 { - let v = v.replace(1, 1 as $elem_ty); - assert_eq!(v.max_element(), 1 as $elem_ty); - } - let v = v.replace(0, 2 as $elem_ty); - assert_eq!(v.max_element(), 2 as $elem_ty); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - pub fn min_element() { - let v = $id::splat(0 as $elem_ty); - assert_eq!(v.min_element(), 0 as $elem_ty); - if $id::lanes() > 1 { - let v = v.replace(1, 1 as $elem_ty); - assert_eq!(v.min_element(), 0 as $elem_ty); - } - let v = $id::splat(1 as $elem_ty); - let v = v.replace(0, 2 as $elem_ty); - if $id::lanes() > 1 { - assert_eq!(v.min_element(), 1 as $elem_ty); - } else { - assert_eq!(v.min_element(), 2 as $elem_ty); - } - if $id::lanes() > 1 { - let v = $id::splat(2 as $elem_ty); - let v = v.replace(1, 1 as $elem_ty); - assert_eq!(v.min_element(), 1 as $elem_ty); - } - } - } - } - } - }; -} - -macro_rules! test_reduction_float_min_max { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - test_if! { - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _reduction_min_max_nan>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn min_element_test() { - let n = crate::$elem_ty::NAN; - - assert_eq!(n.min(-3.), -3.); - assert_eq!((-3. as $elem_ty).min(n), -3.); - - let v0 = $id::splat(-3.); - - let target_with_broken_last_lane_nan = !cfg!(any( - target_arch = "arm", target_arch = "aarch64", - all(target_arch = "x86", - not(target_feature = "sse2") - ), - target_arch = "powerpc64", - target_arch = "wasm32", - )); - - // The vector is initialized to `-3.`s: [-3, -3, -3, -3] - for i in 0..$id::lanes() { - // We replace the i-th element of the vector with - // `NaN`: [-3, -3, -3, NaN] - let mut v = v0.replace(i, n); - - // If the NaN is in the last place, the LLVM - // implementation of these methods is broken on some - // targets: - if i == $id::lanes() - 1 && - target_with_broken_last_lane_nan { - assert_eq!(v.min_element(), -3., - "[A]: nan at {} => {} | {:?}", - i, v.min_element(), v); - - // If we replace all the elements in the vector - // up-to the `i-th` lane with `NaN`s, the result - // is still always `-3.` unless all elements of - // the vector are `NaN`s: - for j in 0..i { - v = v.replace(j, n); - if j == i-1 { - assert!(v.min_element().is_nan(), - "[B]: nan at {} => {} | {:?}", - i, v.min_element(), v); - } else { - assert_eq!(v.min_element(), -3., - "[B]: nan at {} => {} | {:?}", - i, v.min_element(), v); - } - } - - // We are done here, since we were in the last - // lane which is the last iteration of the loop. - break - } - - // We are not in the last lane, and there is only - // one `NaN` in the vector. - - // If the vector has one lane, the result is `NaN`: - if $id::lanes() == 1 { - assert!(v.min_element().is_nan(), - "[C]: all nans | v={:?} | min={} | \ -is_nan: {}", - v, v.min_element(), - v.min_element().is_nan() - ); - - // And we are done, since the vector only has - // one lane anyways. - break; - } - - // The vector has more than one lane, since there is - // only one `NaN` in the vector, the result is - // always `-3`. - assert_eq!(v.min_element(), -3., - "[D]: nan at {} => {} | {:?}", - i, v.min_element(), v); - - // If we replace all the elements in the vector - // up-to the `i-th` lane with `NaN`s, the result is - // still always `-3.` unless all elements of the - // vector are `NaN`s: - for j in 0..i { - v = v.replace(j, n); - - if i == $id::lanes() - 1 && j == i - 1 { - // All elements of the vector are `NaN`s, - // therefore the result is NaN as well. - // - // Note: the #lanes of the vector is > 1, so - // "i - 1" does not overflow. - assert!(v.min_element().is_nan(), - "[E]: all nans | v={:?} | min={} | \ -is_nan: {}", - v, v.min_element(), - v.min_element().is_nan()); - } else { - // There are non-`NaN` elements in the - // vector, therefore the result is `-3.`: - assert_eq!(v.min_element(), -3., - "[F]: nan at {} => {} | {:?}", - i, v.min_element(), v); - } - } - } - - // If the vector contains all NaNs the result is NaN: - assert!($id::splat(n).min_element().is_nan(), - "all nans | v={:?} | min={} | is_nan: {}", - $id::splat(n), $id::splat(n).min_element(), - $id::splat(n).min_element().is_nan()); - } - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn max_element_test() { - let n = crate::$elem_ty::NAN; - - assert_eq!(n.max(-3.), -3.); - assert_eq!((-3. as $elem_ty).max(n), -3.); - - let v0 = $id::splat(-3.); - - let target_with_broken_last_lane_nan = !cfg!(any( - target_arch = "arm", target_arch = "aarch64", - target_arch = "powerpc64", target_arch = "wasm32", - )); - - // The vector is initialized to `-3.`s: [-3, -3, -3, -3] - for i in 0..$id::lanes() { - // We replace the i-th element of the vector with - // `NaN`: [-3, -3, -3, NaN] - let mut v = v0.replace(i, n); - - // If the NaN is in the last place, the LLVM - // implementation of these methods is broken on some - // targets: - if i == $id::lanes() - 1 && - target_with_broken_last_lane_nan { - assert_eq!(v.max_element(), -3., - "[A]: nan at {} => {} | {:?}", - i, v.max_element(), v); - - // If we replace all the elements in the vector - // up-to the `i-th` lane with `NaN`s, the result - // is still always `-3.` unless all elements of - // the vector are `NaN`s: - for j in 0..i { - v = v.replace(j, n); - if j == i-1 { - assert!(v.min_element().is_nan(), - "[B]: nan at {} => {} | {:?}", - i, v.min_element(), v); - } else { - assert_eq!(v.max_element(), -3., - "[B]: nan at {} => {} | {:?}", - i, v.max_element(), v); - } - } - - // We are done here, since we were in the last - // lane which is the last iteration of the loop. - break - } - - // We are not in the last lane, and there is only - // one `NaN` in the vector. - - // If the vector has one lane, the result is `NaN`: - if $id::lanes() == 1 { - assert!(v.max_element().is_nan(), - "[C]: all nans | v={:?} | min={} | \ -is_nan: {}", - v, v.max_element(), - v.max_element().is_nan()); - - // And we are done, since the vector only has - // one lane anyways. - break; - } - - // The vector has more than one lane, since there is - // only one `NaN` in the vector, the result is - // always `-3`. - assert_eq!(v.max_element(), -3., - "[D]: nan at {} => {} | {:?}", - i, v.max_element(), v); - - // If we replace all the elements in the vector - // up-to the `i-th` lane with `NaN`s, the result is - // still always `-3.` unless all elements of the - // vector are `NaN`s: - for j in 0..i { - v = v.replace(j, n); - - if i == $id::lanes() - 1 && j == i - 1 { - // All elements of the vector are `NaN`s, - // therefore the result is NaN as well. - // - // Note: the #lanes of the vector is > 1, so - // "i - 1" does not overflow. - assert!(v.max_element().is_nan(), - "[E]: all nans | v={:?} | max={} | \ -is_nan: {}", - v, v.max_element(), - v.max_element().is_nan()); - } else { - // There are non-`NaN` elements in the - // vector, therefore the result is `-3.`: - assert_eq!(v.max_element(), -3., - "[F]: nan at {} => {} | {:?}", - i, v.max_element(), v); - } - } - } - - // If the vector contains all NaNs the result is NaN: - assert!($id::splat(n).max_element().is_nan(), - "all nans | v={:?} | max={} | is_nan: {}", - $id::splat(n), $id::splat(n).max_element(), - $id::splat(n).max_element().is_nan()); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/select.rs b/third_party/rust/packed_simd/src/api/select.rs deleted file mode 100644 index daf6294721..0000000000 --- a/third_party/rust/packed_simd/src/api/select.rs +++ /dev/null @@ -1,73 +0,0 @@ -//! Implements mask's `select`. - -/// Implements mask select method -macro_rules! impl_select { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Selects elements of `a` and `b` using mask. - /// - /// The lanes of the result for which the mask is `true` contain - /// the values of `a`. The remaining lanes contain the values of - /// `b`. - #[inline] - pub fn select<T>(self, a: Simd<T>, b: Simd<T>) -> Simd<T> - where - T: sealed::SimdArray<NT = <[$elem_ty; $elem_count] as sealed::SimdArray>::NT>, - { - use crate::llvm::simd_select; - Simd(unsafe { simd_select(self.0, a.0, b.0) }) - } - } - - test_select!(bool, $id, $id, (false, true) | $test_tt); - }; -} - -macro_rules! test_select { - ( - $elem_ty:ident, - $mask_ty:ident, - $vec_ty:ident,($small:expr, $large:expr) | - $test_tt:tt - ) => { - test_if! { - $test_tt: - paste::item! { - pub mod [<$vec_ty _select>] { - use super::*; - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn select() { - let o = $small as $elem_ty; - let t = $large as $elem_ty; - - let a = $vec_ty::splat(o); - let b = $vec_ty::splat(t); - let m = a.lt(b); - assert_eq!(m.select(a, b), a); - - let m = b.lt(a); - assert_eq!(m.select(b, a), a); - - let mut c = a; - let mut d = b; - let mut m_e = $mask_ty::splat(false); - for i in 0..$vec_ty::lanes() { - if i % 2 == 0 { - let c_tmp = c.extract(i); - c = c.replace(i, d.extract(i)); - d = d.replace(i, c_tmp); - } else { - m_e = m_e.replace(i, true); - } - } - - let m = c.lt(d); - assert_eq!(m_e, m); - assert_eq!(m.select(c, d), a); - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/shuffle.rs b/third_party/rust/packed_simd/src/api/shuffle.rs deleted file mode 100644 index 1c17bd766e..0000000000 --- a/third_party/rust/packed_simd/src/api/shuffle.rs +++ /dev/null @@ -1,184 +0,0 @@ -//! Implements portable vector shuffles with immediate indices. - -// FIXME: comprehensive tests -// https://github.com/rust-lang-nursery/packed_simd/issues/20 - -/// Shuffles vector elements. -/// -/// This macro returns a new vector that contains a shuffle of the elements in -/// one (`shuffle!(vec, [indices...])`) or two (`shuffle!(vec0, vec1, -/// [indices...])`) input vectors. -/// -/// The type of `vec0` and `vec1` must be equal, and the element type of the -/// resulting vector is the element type of the input vector. -/// -/// The number of `indices` must be a power-of-two in range `[0, 64)`, since -/// currently, the largest vector supported by the library has 64 lanes. The -/// length of the resulting vector equals the number of indices provided. -/// -/// The indices must be in range `[0, M * N)` where `M` is the number of input -/// vectors (`1` or `2`) and `N` is the number of lanes of the input vectors. -/// The indices `i` in range `[0, N)` refer to the `i`-th element of `vec0`, -/// while the indices in range `[N, 2*N)` refer to the `i - N`-th element of -/// `vec1`. -/// -/// # Examples -/// -/// Shuffling elements of two vectors: -/// -/// ``` -/// # use packed_simd::*; -/// # fn main() { -/// // Shuffle allows reordering the elements: -/// let x = i32x4::new(1, 2, 3, 4); -/// let y = i32x4::new(5, 6, 7, 8); -/// let r = shuffle!(x, y, [4, 0, 5, 1]); -/// assert_eq!(r, i32x4::new(5, 1, 6, 2)); -/// -/// // The resulting vector can als be smaller than the input: -/// let r = shuffle!(x, y, [1, 6]); -/// assert_eq!(r, i32x2::new(2, 7)); -/// -/// // Or larger: -/// let r = shuffle!(x, y, [1, 3, 4, 2, 1, 7, 2, 2]); -/// assert_eq!(r, i32x8::new(2, 4, 5, 3, 2, 8, 3, 3)); -/// // At most 2 * the number of lanes in the input vector. -/// # } -/// ``` -/// -/// Shuffling elements of one vector: -/// -/// ``` -/// # use packed_simd::*; -/// # fn main() { -/// // Shuffle allows reordering the elements of a vector: -/// let x = i32x4::new(1, 2, 3, 4); -/// let r = shuffle!(x, [2, 1, 3, 0]); -/// assert_eq!(r, i32x4::new(3, 2, 4, 1)); -/// -/// // The resulting vector can be smaller than the input: -/// let r = shuffle!(x, [1, 3]); -/// assert_eq!(r, i32x2::new(2, 4)); -/// -/// // Equal: -/// let r = shuffle!(x, [1, 3, 2, 0]); -/// assert_eq!(r, i32x4::new(2, 4, 3, 1)); -/// -/// // Or larger: -/// let r = shuffle!(x, [1, 3, 2, 2, 1, 3, 2, 2]); -/// assert_eq!(r, i32x8::new(2, 4, 3, 3, 2, 4, 3, 3)); -/// // At most 2 * the number of lanes in the input vector. -/// # } -/// ``` -#[macro_export] -macro_rules! shuffle { - ($vec0:expr, $vec1:expr, [$l0:expr, $l1:expr]) => {{ - #[allow(unused_unsafe)] - unsafe { - $crate::Simd($crate::__shuffle_vector2::<{[$l0, $l1]}, _, _>( - $vec0.0, - $vec1.0, - )) - } - }}; - ($vec0:expr, $vec1:expr, [$l0:expr, $l1:expr, $l2:expr, $l3:expr]) => {{ - #[allow(unused_unsafe)] - unsafe { - $crate::Simd($crate::__shuffle_vector4::<{[$l0, $l1, $l2, $l3]}, _, _>( - $vec0.0, - $vec1.0, - )) - } - }}; - ($vec0:expr, $vec1:expr, - [$l0:expr, $l1:expr, $l2:expr, $l3:expr, - $l4:expr, $l5:expr, $l6:expr, $l7:expr]) => {{ - #[allow(unused_unsafe)] - unsafe { - $crate::Simd($crate::__shuffle_vector8::<{[$l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7]}, _, _>( - $vec0.0, - $vec1.0, - )) - } - }}; - ($vec0:expr, $vec1:expr, - [$l0:expr, $l1:expr, $l2:expr, $l3:expr, - $l4:expr, $l5:expr, $l6:expr, $l7:expr, - $l8:expr, $l9:expr, $l10:expr, $l11:expr, - $l12:expr, $l13:expr, $l14:expr, $l15:expr]) => {{ - #[allow(unused_unsafe)] - unsafe { - $crate::Simd($crate::__shuffle_vector16::<{ - [ - $l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7, $l8, $l9, $l10, - $l11, $l12, $l13, $l14, $l15, - ] - }, _, _>( - $vec0.0, - $vec1.0, - )) - } - }}; - ($vec0:expr, $vec1:expr, - [$l0:expr, $l1:expr, $l2:expr, $l3:expr, - $l4:expr, $l5:expr, $l6:expr, $l7:expr, - $l8:expr, $l9:expr, $l10:expr, $l11:expr, - $l12:expr, $l13:expr, $l14:expr, $l15:expr, - $l16:expr, $l17:expr, $l18:expr, $l19:expr, - $l20:expr, $l21:expr, $l22:expr, $l23:expr, - $l24:expr, $l25:expr, $l26:expr, $l27:expr, - $l28:expr, $l29:expr, $l30:expr, $l31:expr]) => {{ - #[allow(unused_unsafe)] - unsafe { - $crate::Simd($crate::__shuffle_vector32::<{ - [ - $l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7, $l8, $l9, $l10, - $l11, $l12, $l13, $l14, $l15, $l16, $l17, $l18, $l19, - $l20, $l21, $l22, $l23, $l24, $l25, $l26, $l27, $l28, - $l29, $l30, $l31, - ] - }, _, _>( - $vec0.0, - $vec1.0, - )) - } - }}; - ($vec0:expr, $vec1:expr, - [$l0:expr, $l1:expr, $l2:expr, $l3:expr, - $l4:expr, $l5:expr, $l6:expr, $l7:expr, - $l8:expr, $l9:expr, $l10:expr, $l11:expr, - $l12:expr, $l13:expr, $l14:expr, $l15:expr, - $l16:expr, $l17:expr, $l18:expr, $l19:expr, - $l20:expr, $l21:expr, $l22:expr, $l23:expr, - $l24:expr, $l25:expr, $l26:expr, $l27:expr, - $l28:expr, $l29:expr, $l30:expr, $l31:expr, - $l32:expr, $l33:expr, $l34:expr, $l35:expr, - $l36:expr, $l37:expr, $l38:expr, $l39:expr, - $l40:expr, $l41:expr, $l42:expr, $l43:expr, - $l44:expr, $l45:expr, $l46:expr, $l47:expr, - $l48:expr, $l49:expr, $l50:expr, $l51:expr, - $l52:expr, $l53:expr, $l54:expr, $l55:expr, - $l56:expr, $l57:expr, $l58:expr, $l59:expr, - $l60:expr, $l61:expr, $l62:expr, $l63:expr]) => {{ - #[allow(unused_unsafe)] - unsafe { - $crate::Simd($crate::__shuffle_vector64::<{[ - $l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7, $l8, $l9, $l10, - $l11, $l12, $l13, $l14, $l15, $l16, $l17, $l18, $l19, - $l20, $l21, $l22, $l23, $l24, $l25, $l26, $l27, $l28, - $l29, $l30, $l31, $l32, $l33, $l34, $l35, $l36, $l37, - $l38, $l39, $l40, $l41, $l42, $l43, $l44, $l45, $l46, - $l47, $l48, $l49, $l50, $l51, $l52, $l53, $l54, $l55, - $l56, $l57, $l58, $l59, $l60, $l61, $l62, $l63, - ]}, _, _>( - $vec0.0, - $vec1.0, - )) - } - }}; - ($vec:expr, [$($l:expr),*]) => { - match $vec { - v => shuffle!(v, v, [$($l),*]) - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/shuffle1_dyn.rs b/third_party/rust/packed_simd/src/api/shuffle1_dyn.rs deleted file mode 100644 index 64536be6cb..0000000000 --- a/third_party/rust/packed_simd/src/api/shuffle1_dyn.rs +++ /dev/null @@ -1,159 +0,0 @@ -//! Shuffle vector elements according to a dynamic vector of indices. - -macro_rules! impl_shuffle1_dyn { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Shuffle vector elements according to `indices`. - #[inline] - pub fn shuffle1_dyn<I>(self, indices: I) -> Self - where - Self: codegen::shuffle1_dyn::Shuffle1Dyn<Indices = I>, - { - codegen::shuffle1_dyn::Shuffle1Dyn::shuffle1_dyn(self, indices) - } - } - }; -} - -macro_rules! test_shuffle1_dyn { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _shuffle1_dyn>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn shuffle1_dyn() { - let increasing = { - let mut v = $id::splat(0 as $elem_ty); - for i in 0..$id::lanes() { - v = v.replace(i, i as $elem_ty); - } - v - }; - let decreasing = { - let mut v = $id::splat(0 as $elem_ty); - for i in 0..$id::lanes() { - v = v.replace( - i, - ($id::lanes() - 1 - i) as $elem_ty - ); - } - v - }; - - type Indices = < - $id as codegen::shuffle1_dyn::Shuffle1Dyn - >::Indices; - let increasing_ids: Indices = increasing.cast(); - let decreasing_ids: Indices = decreasing.cast(); - - assert_eq!( - increasing.shuffle1_dyn(increasing_ids), - increasing, - "(i,i)=>i" - ); - assert_eq!( - decreasing.shuffle1_dyn(increasing_ids), - decreasing, - "(d,i)=>d" - ); - assert_eq!( - increasing.shuffle1_dyn(decreasing_ids), - decreasing, - "(i,d)=>d" - ); - assert_eq!( - decreasing.shuffle1_dyn(decreasing_ids), - increasing, - "(d,d)=>i" - ); - - for i in 0..$id::lanes() { - let v_ids: Indices - = $id::splat(i as $elem_ty).cast(); - assert_eq!(increasing.shuffle1_dyn(v_ids), - $id::splat(increasing.extract(i)) - ); - assert_eq!(decreasing.shuffle1_dyn(v_ids), - $id::splat(decreasing.extract(i)) - ); - assert_eq!( - $id::splat(i as $elem_ty) - .shuffle1_dyn(increasing_ids), - $id::splat(i as $elem_ty) - ); - assert_eq!( - $id::splat(i as $elem_ty) - .shuffle1_dyn(decreasing_ids), - $id::splat(i as $elem_ty) - ); - } - } - } - } - } - }; -} - -macro_rules! test_shuffle1_dyn_mask { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _shuffle1_dyn>] { - use super::*; - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn shuffle1_dyn() { - // alternating = [true, false, true, false, ...] - let mut alternating = $id::splat(false); - for i in 0..$id::lanes() { - if i % 2 == 0 { - alternating = alternating.replace(i, true); - } - } - - type Indices = < - $id as codegen::shuffle1_dyn::Shuffle1Dyn - >::Indices; - // even = [0, 0, 2, 2, 4, 4, ..] - let even = { - let mut v = Indices::splat(0); - for i in 0..$id::lanes() { - if i % 2 == 0 { - v = v.replace(i, (i as u8).into()); - } else { - v = v.replace(i, (i as u8 - 1).into()); - } - } - v - }; - // odd = [1, 1, 3, 3, 5, 5, ...] - let odd = { - let mut v = Indices::splat(0); - for i in 0..$id::lanes() { - if i % 2 != 0 { - v = v.replace(i, (i as u8).into()); - } else { - v = v.replace(i, (i as u8 + 1).into()); - } - } - v - }; - - assert_eq!( - alternating.shuffle1_dyn(even), - $id::splat(true) - ); - if $id::lanes() > 1 { - assert_eq!( - alternating.shuffle1_dyn(odd), - $id::splat(false) - ); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/slice.rs b/third_party/rust/packed_simd/src/api/slice.rs deleted file mode 100644 index 526b848b5c..0000000000 --- a/third_party/rust/packed_simd/src/api/slice.rs +++ /dev/null @@ -1,7 +0,0 @@ -//! Slice from/to methods - -#[macro_use] -mod from_slice; - -#[macro_use] -mod write_to_slice; diff --git a/third_party/rust/packed_simd/src/api/slice/from_slice.rs b/third_party/rust/packed_simd/src/api/slice/from_slice.rs deleted file mode 100644 index cafd6f8213..0000000000 --- a/third_party/rust/packed_simd/src/api/slice/from_slice.rs +++ /dev/null @@ -1,202 +0,0 @@ -//! Implements methods to read a vector type from a slice. - -macro_rules! impl_slice_from_slice { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned - /// to an `align_of::<Self>()` boundary. - #[inline] - pub fn from_slice_aligned(slice: &[$elem_ty]) -> Self { - unsafe { - assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_ptr(); - assert_eq!(target_ptr.align_offset(crate::mem::align_of::<Self>()), 0); - Self::from_slice_aligned_unchecked(slice) - } - } - - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()`. - #[inline] - pub fn from_slice_unaligned(slice: &[$elem_ty]) -> Self { - unsafe { - assert!(slice.len() >= $elem_count); - Self::from_slice_unaligned_unchecked(slice) - } - } - - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned - /// to an `align_of::<Self>()` boundary, the behavior is undefined. - #[inline] - pub unsafe fn from_slice_aligned_unchecked(slice: &[$elem_ty]) -> Self { - debug_assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_ptr(); - debug_assert_eq!(target_ptr.align_offset(crate::mem::align_of::<Self>()), 0); - - #[allow(clippy::cast_ptr_alignment)] - *(target_ptr as *const Self) - } - - /// Instantiates a new vector with the values of the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn from_slice_unaligned_unchecked(slice: &[$elem_ty]) -> Self { - use crate::mem::size_of; - debug_assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_ptr().cast(); - let mut x = Self::splat(0 as $elem_ty); - let self_ptr = &mut x as *mut Self as *mut u8; - crate::ptr::copy_nonoverlapping(target_ptr, self_ptr, size_of::<Self>()); - x - } - } - - test_if! { - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _slice_from_slice>] { - use super::*; - use crate::iter::Iterator; - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from_slice_unaligned() { - let mut unaligned = [42 as $elem_ty; $id::lanes() + 1]; - unaligned[0] = 0 as $elem_ty; - let vec = $id::from_slice_unaligned(&unaligned[1..]); - for (index, &b) in unaligned.iter().enumerate() { - if index == 0 { - assert_eq!(b, 0 as $elem_ty); - } else { - assert_eq!(b, 42 as $elem_ty); - assert_eq!(b, vec.extract(index - 1)); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn from_slice_unaligned_fail() { - let mut unaligned = [42 as $elem_ty; $id::lanes() + 1]; - unaligned[0] = 0 as $elem_ty; - // the slice is not large enough => panic - let _vec = $id::from_slice_unaligned(&unaligned[2..]); - } - - union A { - data: [$elem_ty; 2 * $id::lanes()], - _vec: $id, - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from_slice_aligned() { - let mut aligned = A { - data: [0 as $elem_ty; 2 * $id::lanes()], - }; - for i in $id::lanes()..(2 * $id::lanes()) { - unsafe { - aligned.data[i] = 42 as $elem_ty; - } - } - - let vec = unsafe { - $id::from_slice_aligned( - &aligned.data[$id::lanes()..] - ) - }; - for (index, &b) in - unsafe { aligned.data.iter().enumerate() } { - if index < $id::lanes() { - assert_eq!(b, 0 as $elem_ty); - } else { - assert_eq!(b, 42 as $elem_ty); - assert_eq!( - b, vec.extract(index - $id::lanes()) - ); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn from_slice_aligned_fail_lanes() { - let aligned = A { - data: [0 as $elem_ty; 2 * $id::lanes()], - }; - let _vec = unsafe { - $id::from_slice_aligned( - &aligned.data[2 * $id::lanes()..] - ) - }; - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn from_slice_aligned_fail_align() { - unsafe { - let aligned = A { - data: [0 as $elem_ty; 2 * $id::lanes()], - }; - - // get a pointer to the front of data - let ptr: *const $elem_ty = aligned.data.as_ptr() - as *const $elem_ty; - // offset pointer by one element - let ptr = ptr.wrapping_add(1); - - if ptr.align_offset( - crate::mem::align_of::<$id>() - ) == 0 { - // the pointer is properly aligned, so - // from_slice_aligned won't fail here (e.g. this - // can happen for i128x1). So we panic to make - // the "should_fail" test pass: - panic!("ok"); - } - - // create a slice - this is safe, because the - // elements of the slice exist, are properly - // initialized, and properly aligned: - let s: &[$elem_ty] = slice::from_raw_parts( - ptr, $id::lanes() - ); - // this should always panic because the slice - // alignment does not match the alignment - // requirements for the vector type: - let _vec = $id::from_slice_aligned(s); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/slice/write_to_slice.rs b/third_party/rust/packed_simd/src/api/slice/write_to_slice.rs deleted file mode 100644 index 5abd4916e0..0000000000 --- a/third_party/rust/packed_simd/src/api/slice/write_to_slice.rs +++ /dev/null @@ -1,196 +0,0 @@ -//! Implements methods to write a vector type to a slice. - -macro_rules! impl_slice_write_to_slice { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Writes the values of the vector to the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not - /// aligned to an `align_of::<Self>()` boundary. - #[inline] - pub fn write_to_slice_aligned(self, slice: &mut [$elem_ty]) { - unsafe { - assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_mut_ptr(); - assert_eq!(target_ptr.align_offset(crate::mem::align_of::<Self>()), 0); - self.write_to_slice_aligned_unchecked(slice); - } - } - - /// Writes the values of the vector to the `slice`. - /// - /// # Panics - /// - /// If `slice.len() < Self::lanes()`. - #[inline] - pub fn write_to_slice_unaligned(self, slice: &mut [$elem_ty]) { - unsafe { - assert!(slice.len() >= $elem_count); - self.write_to_slice_unaligned_unchecked(slice); - } - } - - /// Writes the values of the vector to the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` or `&slice[0]` is not - /// aligned to an `align_of::<Self>()` boundary, the behavior is - /// undefined. - #[inline] - pub unsafe fn write_to_slice_aligned_unchecked(self, slice: &mut [$elem_ty]) { - debug_assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_mut_ptr(); - debug_assert_eq!(target_ptr.align_offset(crate::mem::align_of::<Self>()), 0); - - #[allow(clippy::cast_ptr_alignment)] - #[allow(clippy::cast_ptr_alignment)] - #[allow(clippy::cast_ptr_alignment)] - #[allow(clippy::cast_ptr_alignment)] - *(target_ptr as *mut Self) = self; - } - - /// Writes the values of the vector to the `slice`. - /// - /// # Safety - /// - /// If `slice.len() < Self::lanes()` the behavior is undefined. - #[inline] - pub unsafe fn write_to_slice_unaligned_unchecked(self, slice: &mut [$elem_ty]) { - debug_assert!(slice.len() >= $elem_count); - let target_ptr = slice.as_mut_ptr().cast(); - let self_ptr = &self as *const Self as *const u8; - crate::ptr::copy_nonoverlapping(self_ptr, target_ptr, crate::mem::size_of::<Self>()); - } - } - - test_if! { - $test_tt: - paste::item! { - // Comparisons use integer casts within mantissa^1 range. - #[allow(clippy::float_cmp)] - pub mod [<$id _slice_write_to_slice>] { - use super::*; - use crate::iter::Iterator; - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn write_to_slice_unaligned() { - let mut unaligned = [0 as $elem_ty; $id::lanes() + 1]; - let vec = $id::splat(42 as $elem_ty); - vec.write_to_slice_unaligned(&mut unaligned[1..]); - for (index, &b) in unaligned.iter().enumerate() { - if index == 0 { - assert_eq!(b, 0 as $elem_ty); - } else { - assert_eq!(b, 42 as $elem_ty); - assert_eq!(b, vec.extract(index - 1)); - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn write_to_slice_unaligned_fail() { - let mut unaligned = [0 as $elem_ty; $id::lanes() + 1]; - let vec = $id::splat(42 as $elem_ty); - vec.write_to_slice_unaligned(&mut unaligned[2..]); - } - - union A { - data: [$elem_ty; 2 * $id::lanes()], - _vec: $id, - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] - #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn write_to_slice_aligned() { - let mut aligned = A { - data: [0 as $elem_ty; 2 * $id::lanes()], - }; - let vec = $id::splat(42 as $elem_ty); - unsafe { - vec.write_to_slice_aligned( - &mut aligned.data[$id::lanes()..] - ); - for (idx, &b) in aligned.data.iter().enumerate() { - if idx < $id::lanes() { - assert_eq!(b, 0 as $elem_ty); - } else { - assert_eq!(b, 42 as $elem_ty); - assert_eq!( - b, vec.extract(idx - $id::lanes()) - ); - } - } - } - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn write_to_slice_aligned_fail_lanes() { - let mut aligned = A { - data: [0 as $elem_ty; 2 * $id::lanes()], - }; - let vec = $id::splat(42 as $elem_ty); - unsafe { - vec.write_to_slice_aligned( - &mut aligned.data[2 * $id::lanes()..] - ) - }; - } - - // FIXME: wasm-bindgen-test does not support #[should_panic] - // #[cfg_attr(not(target_arch = "wasm32"), test)] - // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - #[cfg(not(target_arch = "wasm32"))] - #[test] - #[should_panic] - fn write_to_slice_aligned_fail_align() { - unsafe { - let mut aligned = A { - data: [0 as $elem_ty; 2 * $id::lanes()], - }; - - // get a pointer to the front of data - let ptr: *mut $elem_ty - = aligned.data.as_mut_ptr() as *mut $elem_ty; - // offset pointer by one element - let ptr = ptr.wrapping_add(1); - - if ptr.align_offset(crate::mem::align_of::<$id>()) - == 0 { - // the pointer is properly aligned, so - // write_to_slice_aligned won't fail here (e.g. - // this can happen for i128x1). So we panic to - // make the "should_fail" test pass: - panic!("ok"); - } - - // create a slice - this is safe, because the - // elements of the slice exist, are properly - // initialized, and properly aligned: - let s: &mut [$elem_ty] - = slice::from_raw_parts_mut(ptr, $id::lanes()); - // this should always panic because the slice - // alignment does not match the alignment - // requirements for the vector type: - let vec = $id::splat(42 as $elem_ty); - vec.write_to_slice_aligned(s); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/api/swap_bytes.rs b/third_party/rust/packed_simd/src/api/swap_bytes.rs deleted file mode 100644 index 4649ed679b..0000000000 --- a/third_party/rust/packed_simd/src/api/swap_bytes.rs +++ /dev/null @@ -1,192 +0,0 @@ -//! Horizontal swap bytes - -macro_rules! impl_swap_bytes { - ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { - impl $id { - /// Reverses the byte order of the vector. - #[inline] - pub fn swap_bytes(self) -> Self { - super::codegen::swap_bytes::SwapBytes::swap_bytes(self) - } - - /// Converts self to little endian from the target's endianness. - /// - /// On little endian this is a no-op. On big endian the bytes are - /// swapped. - #[inline] - pub fn to_le(self) -> Self { - #[cfg(target_endian = "little")] - { - self - } - #[cfg(not(target_endian = "little"))] - { - self.swap_bytes() - } - } - - /// Converts self to big endian from the target's endianness. - /// - /// On big endian this is a no-op. On little endian the bytes are - /// swapped. - #[inline] - pub fn to_be(self) -> Self { - #[cfg(target_endian = "big")] - { - self - } - #[cfg(not(target_endian = "big"))] - { - self.swap_bytes() - } - } - - /// Converts a vector from little endian to the target's endianness. - /// - /// On little endian this is a no-op. On big endian the bytes are - /// swapped. - #[inline] - pub fn from_le(x: Self) -> Self { - #[cfg(target_endian = "little")] - { - x - } - #[cfg(not(target_endian = "little"))] - { - x.swap_bytes() - } - } - - /// Converts a vector from big endian to the target's endianness. - /// - /// On big endian this is a no-op. On little endian the bytes are - /// swapped. - #[inline] - pub fn from_be(x: Self) -> Self { - #[cfg(target_endian = "big")] - { - x - } - #[cfg(not(target_endian = "big"))] - { - x.swap_bytes() - } - } - } - - test_if! { - $test_tt: - paste::item! { - pub mod [<$id _swap_bytes>] { - use super::*; - - const BYTES: [u8; 64] = [ - 0, 1, 2, 3, 4, 5, 6, 7, - 8, 9, 10, 11, 12, 13, 14, 15, - 16, 17, 18, 19, 20, 21, 22, 23, - 24, 25, 26, 27, 28, 29, 30, 31, - 32, 33, 34, 35, 36, 37, 38, 39, - 40, 41, 42, 43, 44, 45, 46, 47, - 48, 49, 50, 51, 52, 53, 54, 55, - 56, 57, 58, 59, 60, 61, 62, 63, - ]; - - macro_rules! swap { - ($func: ident) => {{ - // catch possible future >512 vectors - assert!(mem::size_of::<$id>() <= 64); - - let mut actual = BYTES; - let elems: &mut [$elem_ty] = unsafe { - slice::from_raw_parts_mut( - actual.as_mut_ptr() as *mut $elem_ty, - $id::lanes(), - ) - }; - - let vec = $id::from_slice_unaligned(elems); - $id::$func(vec).write_to_slice_unaligned(elems); - - actual - }}; - } - - macro_rules! test_swap { - ($func: ident) => {{ - let actual = swap!($func); - let expected = - BYTES.iter().rev() - .skip(64 - crate::mem::size_of::<$id>()); - assert!(actual.iter().zip(expected) - .all(|(x, y)| x == y)); - }}; - } - - macro_rules! test_no_swap { - ($func: ident) => {{ - let actual = swap!($func); - let expected = BYTES.iter() - .take(mem::size_of::<$id>()); - - assert!(actual.iter().zip(expected) - .all(|(x, y)| x == y)); - }}; - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn swap_bytes() { - test_swap!(swap_bytes); - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn to_le() { - #[cfg(target_endian = "little")] - { - test_no_swap!(to_le); - } - #[cfg(not(target_endian = "little"))] - { - test_swap!(to_le); - } - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn to_be() { - #[cfg(target_endian = "big")] - { - test_no_swap!(to_be); - } - #[cfg(not(target_endian = "big"))] - { - test_swap!(to_be); - } - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from_le() { - #[cfg(target_endian = "little")] - { - test_no_swap!(from_le); - } - #[cfg(not(target_endian = "little"))] - { - test_swap!(from_le); - } - } - - #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] - fn from_be() { - #[cfg(target_endian = "big")] - { - test_no_swap!(from_be); - } - #[cfg(not(target_endian = "big"))] - { - test_swap!(from_be); - } - } - } - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen.rs b/third_party/rust/packed_simd/src/codegen.rs deleted file mode 100644 index 8a9e971486..0000000000 --- a/third_party/rust/packed_simd/src/codegen.rs +++ /dev/null @@ -1,62 +0,0 @@ -//! Code-generation utilities - -pub(crate) mod bit_manip; -pub(crate) mod llvm; -pub(crate) mod math; -pub(crate) mod reductions; -pub(crate) mod shuffle; -pub(crate) mod shuffle1_dyn; -pub(crate) mod swap_bytes; - -macro_rules! impl_simd_array { - ([$elem_ty:ident; $elem_count:expr]: - $tuple_id:ident | $($elem_tys:ident),*) => { - #[derive(Copy, Clone)] - #[repr(simd)] - pub struct $tuple_id($(pub(crate) $elem_tys),*); - //^^^^^^^ leaked through SimdArray - - impl crate::sealed::Seal for [$elem_ty; $elem_count] {} - - impl crate::sealed::SimdArray for [$elem_ty; $elem_count] { - type Tuple = $tuple_id; - type T = $elem_ty; - const N: usize = $elem_count; - type NT = [u32; $elem_count]; - } - - impl crate::sealed::Seal for $tuple_id {} - impl crate::sealed::Simd for $tuple_id { - type Element = $elem_ty; - const LANES: usize = $elem_count; - type LanesType = [u32; $elem_count]; - } - - } -} - -pub(crate) mod pointer_sized_int; - -pub(crate) mod v16; -pub(crate) use self::v16::*; - -pub(crate) mod v32; -pub(crate) use self::v32::*; - -pub(crate) mod v64; -pub(crate) use self::v64::*; - -pub(crate) mod v128; -pub(crate) use self::v128::*; - -pub(crate) mod v256; -pub(crate) use self::v256::*; - -pub(crate) mod v512; -pub(crate) use self::v512::*; - -pub(crate) mod vSize; -pub(crate) use self::vSize::*; - -pub(crate) mod vPtr; -pub(crate) use self::vPtr::*; diff --git a/third_party/rust/packed_simd/src/codegen/bit_manip.rs b/third_party/rust/packed_simd/src/codegen/bit_manip.rs deleted file mode 100644 index 32d8d717a0..0000000000 --- a/third_party/rust/packed_simd/src/codegen/bit_manip.rs +++ /dev/null @@ -1,347 +0,0 @@ -//! LLVM bit manipulation intrinsics. -#[rustfmt::skip] - -pub(crate) use crate::*; - -#[allow(improper_ctypes, dead_code)] -extern "C" { - #[link_name = "llvm.ctlz.v2i8"] - fn ctlz_u8x2(x: u8x2, is_zero_undef: bool) -> u8x2; - #[link_name = "llvm.ctlz.v4i8"] - fn ctlz_u8x4(x: u8x4, is_zero_undef: bool) -> u8x4; - #[link_name = "llvm.ctlz.v8i8"] - fn ctlz_u8x8(x: u8x8, is_zero_undef: bool) -> u8x8; - #[link_name = "llvm.ctlz.v16i8"] - fn ctlz_u8x16(x: u8x16, is_zero_undef: bool) -> u8x16; - #[link_name = "llvm.ctlz.v32i8"] - fn ctlz_u8x32(x: u8x32, is_zero_undef: bool) -> u8x32; - #[link_name = "llvm.ctlz.v64i8"] - fn ctlz_u8x64(x: u8x64, is_zero_undef: bool) -> u8x64; - - #[link_name = "llvm.ctlz.v2i16"] - fn ctlz_u16x2(x: u16x2, is_zero_undef: bool) -> u16x2; - #[link_name = "llvm.ctlz.v4i16"] - fn ctlz_u16x4(x: u16x4, is_zero_undef: bool) -> u16x4; - #[link_name = "llvm.ctlz.v8i16"] - fn ctlz_u16x8(x: u16x8, is_zero_undef: bool) -> u16x8; - #[link_name = "llvm.ctlz.v16i16"] - fn ctlz_u16x16(x: u16x16, is_zero_undef: bool) -> u16x16; - #[link_name = "llvm.ctlz.v32i16"] - fn ctlz_u16x32(x: u16x32, is_zero_undef: bool) -> u16x32; - - #[link_name = "llvm.ctlz.v2i32"] - fn ctlz_u32x2(x: u32x2, is_zero_undef: bool) -> u32x2; - #[link_name = "llvm.ctlz.v4i32"] - fn ctlz_u32x4(x: u32x4, is_zero_undef: bool) -> u32x4; - #[link_name = "llvm.ctlz.v8i32"] - fn ctlz_u32x8(x: u32x8, is_zero_undef: bool) -> u32x8; - #[link_name = "llvm.ctlz.v16i32"] - fn ctlz_u32x16(x: u32x16, is_zero_undef: bool) -> u32x16; - - #[link_name = "llvm.ctlz.v2i64"] - fn ctlz_u64x2(x: u64x2, is_zero_undef: bool) -> u64x2; - #[link_name = "llvm.ctlz.v4i64"] - fn ctlz_u64x4(x: u64x4, is_zero_undef: bool) -> u64x4; - #[link_name = "llvm.ctlz.v8i64"] - fn ctlz_u64x8(x: u64x8, is_zero_undef: bool) -> u64x8; - - #[link_name = "llvm.ctlz.v1i128"] - fn ctlz_u128x1(x: u128x1, is_zero_undef: bool) -> u128x1; - #[link_name = "llvm.ctlz.v2i128"] - fn ctlz_u128x2(x: u128x2, is_zero_undef: bool) -> u128x2; - #[link_name = "llvm.ctlz.v4i128"] - fn ctlz_u128x4(x: u128x4, is_zero_undef: bool) -> u128x4; - - #[link_name = "llvm.cttz.v2i8"] - fn cttz_u8x2(x: u8x2, is_zero_undef: bool) -> u8x2; - #[link_name = "llvm.cttz.v4i8"] - fn cttz_u8x4(x: u8x4, is_zero_undef: bool) -> u8x4; - #[link_name = "llvm.cttz.v8i8"] - fn cttz_u8x8(x: u8x8, is_zero_undef: bool) -> u8x8; - #[link_name = "llvm.cttz.v16i8"] - fn cttz_u8x16(x: u8x16, is_zero_undef: bool) -> u8x16; - #[link_name = "llvm.cttz.v32i8"] - fn cttz_u8x32(x: u8x32, is_zero_undef: bool) -> u8x32; - #[link_name = "llvm.cttz.v64i8"] - fn cttz_u8x64(x: u8x64, is_zero_undef: bool) -> u8x64; - - #[link_name = "llvm.cttz.v2i16"] - fn cttz_u16x2(x: u16x2, is_zero_undef: bool) -> u16x2; - #[link_name = "llvm.cttz.v4i16"] - fn cttz_u16x4(x: u16x4, is_zero_undef: bool) -> u16x4; - #[link_name = "llvm.cttz.v8i16"] - fn cttz_u16x8(x: u16x8, is_zero_undef: bool) -> u16x8; - #[link_name = "llvm.cttz.v16i16"] - fn cttz_u16x16(x: u16x16, is_zero_undef: bool) -> u16x16; - #[link_name = "llvm.cttz.v32i16"] - fn cttz_u16x32(x: u16x32, is_zero_undef: bool) -> u16x32; - - #[link_name = "llvm.cttz.v2i32"] - fn cttz_u32x2(x: u32x2, is_zero_undef: bool) -> u32x2; - #[link_name = "llvm.cttz.v4i32"] - fn cttz_u32x4(x: u32x4, is_zero_undef: bool) -> u32x4; - #[link_name = "llvm.cttz.v8i32"] - fn cttz_u32x8(x: u32x8, is_zero_undef: bool) -> u32x8; - #[link_name = "llvm.cttz.v16i32"] - fn cttz_u32x16(x: u32x16, is_zero_undef: bool) -> u32x16; - - #[link_name = "llvm.cttz.v2i64"] - fn cttz_u64x2(x: u64x2, is_zero_undef: bool) -> u64x2; - #[link_name = "llvm.cttz.v4i64"] - fn cttz_u64x4(x: u64x4, is_zero_undef: bool) -> u64x4; - #[link_name = "llvm.cttz.v8i64"] - fn cttz_u64x8(x: u64x8, is_zero_undef: bool) -> u64x8; - - #[link_name = "llvm.cttz.v1i128"] - fn cttz_u128x1(x: u128x1, is_zero_undef: bool) -> u128x1; - #[link_name = "llvm.cttz.v2i128"] - fn cttz_u128x2(x: u128x2, is_zero_undef: bool) -> u128x2; - #[link_name = "llvm.cttz.v4i128"] - fn cttz_u128x4(x: u128x4, is_zero_undef: bool) -> u128x4; - - #[link_name = "llvm.ctpop.v2i8"] - fn ctpop_u8x2(x: u8x2) -> u8x2; - #[link_name = "llvm.ctpop.v4i8"] - fn ctpop_u8x4(x: u8x4) -> u8x4; - #[link_name = "llvm.ctpop.v8i8"] - fn ctpop_u8x8(x: u8x8) -> u8x8; - #[link_name = "llvm.ctpop.v16i8"] - fn ctpop_u8x16(x: u8x16) -> u8x16; - #[link_name = "llvm.ctpop.v32i8"] - fn ctpop_u8x32(x: u8x32) -> u8x32; - #[link_name = "llvm.ctpop.v64i8"] - fn ctpop_u8x64(x: u8x64) -> u8x64; - - #[link_name = "llvm.ctpop.v2i16"] - fn ctpop_u16x2(x: u16x2) -> u16x2; - #[link_name = "llvm.ctpop.v4i16"] - fn ctpop_u16x4(x: u16x4) -> u16x4; - #[link_name = "llvm.ctpop.v8i16"] - fn ctpop_u16x8(x: u16x8) -> u16x8; - #[link_name = "llvm.ctpop.v16i16"] - fn ctpop_u16x16(x: u16x16) -> u16x16; - #[link_name = "llvm.ctpop.v32i16"] - fn ctpop_u16x32(x: u16x32) -> u16x32; - - #[link_name = "llvm.ctpop.v2i32"] - fn ctpop_u32x2(x: u32x2) -> u32x2; - #[link_name = "llvm.ctpop.v4i32"] - fn ctpop_u32x4(x: u32x4) -> u32x4; - #[link_name = "llvm.ctpop.v8i32"] - fn ctpop_u32x8(x: u32x8) -> u32x8; - #[link_name = "llvm.ctpop.v16i32"] - fn ctpop_u32x16(x: u32x16) -> u32x16; - - #[link_name = "llvm.ctpop.v2i64"] - fn ctpop_u64x2(x: u64x2) -> u64x2; - #[link_name = "llvm.ctpop.v4i64"] - fn ctpop_u64x4(x: u64x4) -> u64x4; - #[link_name = "llvm.ctpop.v8i64"] - fn ctpop_u64x8(x: u64x8) -> u64x8; - - #[link_name = "llvm.ctpop.v1i128"] - fn ctpop_u128x1(x: u128x1) -> u128x1; - #[link_name = "llvm.ctpop.v2i128"] - fn ctpop_u128x2(x: u128x2) -> u128x2; - #[link_name = "llvm.ctpop.v4i128"] - fn ctpop_u128x4(x: u128x4) -> u128x4; -} - -pub(crate) trait BitManip { - fn ctpop(self) -> Self; - fn ctlz(self) -> Self; - fn cttz(self) -> Self; -} - -macro_rules! impl_bit_manip { - (inner: $ty:ident, $scalar:ty, $uty:ident, - $ctpop:ident, $ctlz:ident, $cttz:ident) => { - // FIXME: several LLVM intrinsics break on s390x https://github.com/rust-lang-nursery/packed_simd/issues/192 - #[cfg(target_arch = "s390x")] - impl_bit_manip! { scalar: $ty, $scalar } - #[cfg(not(target_arch = "s390x"))] - impl BitManip for $ty { - #[inline] - fn ctpop(self) -> Self { - let y: $uty = self.cast(); - unsafe { $ctpop(y).cast() } - } - - #[inline] - fn ctlz(self) -> Self { - let y: $uty = self.cast(); - // the ctxx intrinsics need compile-time constant - // `is_zero_undef` - unsafe { $ctlz(y, false).cast() } - } - - #[inline] - fn cttz(self) -> Self { - let y: $uty = self.cast(); - unsafe { $cttz(y, false).cast() } - } - } - }; - (sized_inner: $ty:ident, $scalar:ty, $uty:ident) => { - #[cfg(target_arch = "s390x")] - impl_bit_manip! { scalar: $ty, $scalar } - #[cfg(not(target_arch = "s390x"))] - impl BitManip for $ty { - #[inline] - fn ctpop(self) -> Self { - let y: $uty = self.cast(); - $uty::ctpop(y).cast() - } - - #[inline] - fn ctlz(self) -> Self { - let y: $uty = self.cast(); - $uty::ctlz(y).cast() - } - - #[inline] - fn cttz(self) -> Self { - let y: $uty = self.cast(); - $uty::cttz(y).cast() - } - } - }; - (scalar: $ty:ident, $scalar:ty) => { - impl BitManip for $ty { - #[inline] - fn ctpop(self) -> Self { - let mut ones = self; - for i in 0..Self::lanes() { - ones = ones.replace(i, self.extract(i).count_ones() as $scalar); - } - ones - } - - #[inline] - fn ctlz(self) -> Self { - let mut lz = self; - for i in 0..Self::lanes() { - lz = lz.replace(i, self.extract(i).leading_zeros() as $scalar); - } - lz - } - - #[inline] - fn cttz(self) -> Self { - let mut tz = self; - for i in 0..Self::lanes() { - tz = tz.replace(i, self.extract(i).trailing_zeros() as $scalar); - } - tz - } - } - }; - ($uty:ident, $uscalar:ty, $ity:ident, $iscalar:ty, - $ctpop:ident, $ctlz:ident, $cttz:ident) => { - impl_bit_manip! { inner: $uty, $uscalar, $uty, $ctpop, $ctlz, $cttz } - impl_bit_manip! { inner: $ity, $iscalar, $uty, $ctpop, $ctlz, $cttz } - }; - (sized: $usize:ident, $uscalar:ty, $isize:ident, - $iscalar:ty, $ty:ident) => { - impl_bit_manip! { sized_inner: $usize, $uscalar, $ty } - impl_bit_manip! { sized_inner: $isize, $iscalar, $ty } - }; -} - -impl_bit_manip! { u8x2 , u8, i8x2, i8, ctpop_u8x2, ctlz_u8x2, cttz_u8x2 } -impl_bit_manip! { u8x4 , u8, i8x4, i8, ctpop_u8x4, ctlz_u8x4, cttz_u8x4 } -#[cfg(not(target_arch = "aarch64"))] // see below -impl_bit_manip! { u8x8 , u8, i8x8, i8, ctpop_u8x8, ctlz_u8x8, cttz_u8x8 } -impl_bit_manip! { u8x16 , u8, i8x16, i8, ctpop_u8x16, ctlz_u8x16, cttz_u8x16 } -impl_bit_manip! { u8x32 , u8, i8x32, i8, ctpop_u8x32, ctlz_u8x32, cttz_u8x32 } -impl_bit_manip! { u8x64 , u8, i8x64, i8, ctpop_u8x64, ctlz_u8x64, cttz_u8x64 } -impl_bit_manip! { u16x2 , u16, i16x2, i16, ctpop_u16x2, ctlz_u16x2, cttz_u16x2 } -impl_bit_manip! { u16x4 , u16, i16x4, i16, ctpop_u16x4, ctlz_u16x4, cttz_u16x4 } -impl_bit_manip! { u16x8 , u16, i16x8, i16, ctpop_u16x8, ctlz_u16x8, cttz_u16x8 } -impl_bit_manip! { u16x16 , u16, i16x16, i16, ctpop_u16x16, ctlz_u16x16, cttz_u16x16 } -impl_bit_manip! { u16x32 , u16, i16x32, i16, ctpop_u16x32, ctlz_u16x32, cttz_u16x32 } -impl_bit_manip! { u32x2 , u32, i32x2, i32, ctpop_u32x2, ctlz_u32x2, cttz_u32x2 } -impl_bit_manip! { u32x4 , u32, i32x4, i32, ctpop_u32x4, ctlz_u32x4, cttz_u32x4 } -impl_bit_manip! { u32x8 , u32, i32x8, i32, ctpop_u32x8, ctlz_u32x8, cttz_u32x8 } -impl_bit_manip! { u32x16 , u32, i32x16, i32, ctpop_u32x16, ctlz_u32x16, cttz_u32x16 } -impl_bit_manip! { u64x2 , u64, i64x2, i64, ctpop_u64x2, ctlz_u64x2, cttz_u64x2 } -impl_bit_manip! { u64x4 , u64, i64x4, i64, ctpop_u64x4, ctlz_u64x4, cttz_u64x4 } -impl_bit_manip! { u64x8 , u64, i64x8, i64, ctpop_u64x8, ctlz_u64x8, cttz_u64x8 } -impl_bit_manip! { u128x1 , u128, i128x1, i128, ctpop_u128x1, ctlz_u128x1, cttz_u128x1 } -impl_bit_manip! { u128x2 , u128, i128x2, i128, ctpop_u128x2, ctlz_u128x2, cttz_u128x2 } -impl_bit_manip! { u128x4 , u128, i128x4, i128, ctpop_u128x4, ctlz_u128x4, cttz_u128x4 } - -#[cfg(target_arch = "aarch64")] -impl BitManip for u8x8 { - #[inline] - fn ctpop(self) -> Self { - let y: u8x8 = self.cast(); - unsafe { ctpop_u8x8(y).cast() } - } - - #[inline] - fn ctlz(self) -> Self { - let y: u8x8 = self.cast(); - unsafe { ctlz_u8x8(y, false).cast() } - } - - #[inline] - fn cttz(self) -> Self { - // FIXME: LLVM cttz.v8i8 broken on aarch64 https://github.com/rust-lang-nursery/packed_simd/issues/191 - // OPTIMIZE: adapt the algorithm used for v8i16/etc to Rust's aarch64 - // intrinsics - let mut tz = self; - for i in 0..Self::lanes() { - tz = tz.replace(i, self.extract(i).trailing_zeros() as u8); - } - tz - } -} -#[cfg(target_arch = "aarch64")] -impl BitManip for i8x8 { - #[inline] - fn ctpop(self) -> Self { - let y: u8x8 = self.cast(); - unsafe { ctpop_u8x8(y).cast() } - } - - #[inline] - fn ctlz(self) -> Self { - let y: u8x8 = self.cast(); - unsafe { ctlz_u8x8(y, false).cast() } - } - - #[inline] - fn cttz(self) -> Self { - // FIXME: LLVM cttz.v8i8 broken on aarch64 https://github.com/rust-lang-nursery/packed_simd/issues/191 - // OPTIMIZE: adapt the algorithm used for v8i16/etc to Rust's aarch64 - // intrinsics - let mut tz = self; - for i in 0..Self::lanes() { - tz = tz.replace(i, self.extract(i).trailing_zeros() as i8); - } - tz - } -} - -cfg_if! { - if #[cfg(target_pointer_width = "8")] { - impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u8x2 } - impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u8x4 } - impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u8x8 } - } else if #[cfg(target_pointer_width = "16")] { - impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u16x2 } - impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u16x4 } - impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u16x8 } - } else if #[cfg(target_pointer_width = "32")] { - impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u32x2 } - impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u32x4 } - impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u32x8 } - } else if #[cfg(target_pointer_width = "64")] { - impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u64x2 } - impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u64x4 } - impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u64x8 } - } else { - compile_error!("unsupported target_pointer_width"); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/llvm.rs b/third_party/rust/packed_simd/src/codegen/llvm.rs deleted file mode 100644 index bb482fac66..0000000000 --- a/third_party/rust/packed_simd/src/codegen/llvm.rs +++ /dev/null @@ -1,122 +0,0 @@ -//! LLVM's platform intrinsics -#![allow(dead_code)] - -use crate::sealed::Shuffle; -#[allow(unused_imports)] // FIXME: spurious warning? -use crate::sealed::Simd; - -extern "platform-intrinsic" { - fn simd_shuffle<T, I, U>(x: T, y: T, idx: I) -> U; -} - -#[allow(clippy::missing_safety_doc)] -#[inline] -pub unsafe fn __shuffle_vector2<const IDX: [u32; 2], T, U>(x: T, y: T) -> U -where - T: Simd, - <T as Simd>::Element: Shuffle<[u32; 2], Output = U>, -{ - simd_shuffle(x, y, IDX) -} - -#[allow(clippy::missing_safety_doc)] -#[inline] -pub unsafe fn __shuffle_vector4<const IDX: [u32; 4], T, U>(x: T, y: T) -> U -where - T: Simd, - <T as Simd>::Element: Shuffle<[u32; 4], Output = U>, -{ - simd_shuffle(x, y, IDX) -} - -#[allow(clippy::missing_safety_doc)] -#[inline] -pub unsafe fn __shuffle_vector8<const IDX: [u32; 8], T, U>(x: T, y: T) -> U -where - T: Simd, - <T as Simd>::Element: Shuffle<[u32; 8], Output = U>, -{ - simd_shuffle(x, y, IDX) -} - -#[allow(clippy::missing_safety_doc)] -#[inline] -pub unsafe fn __shuffle_vector16<const IDX: [u32; 16], T, U>(x: T, y: T) -> U -where - T: Simd, - <T as Simd>::Element: Shuffle<[u32; 16], Output = U>, -{ - simd_shuffle(x, y, IDX) -} - -#[allow(clippy::missing_safety_doc)] -#[inline] -pub unsafe fn __shuffle_vector32<const IDX: [u32; 32], T, U>(x: T, y: T) -> U -where - T: Simd, - <T as Simd>::Element: Shuffle<[u32; 32], Output = U>, -{ - simd_shuffle(x, y, IDX) -} - -#[allow(clippy::missing_safety_doc)] -#[inline] -pub unsafe fn __shuffle_vector64<const IDX: [u32; 64], T, U>(x: T, y: T) -> U -where - T: Simd, - <T as Simd>::Element: Shuffle<[u32; 64], Output = U>, -{ - simd_shuffle(x, y, IDX) -} - -extern "platform-intrinsic" { - pub(crate) fn simd_eq<T, U>(x: T, y: T) -> U; - pub(crate) fn simd_ne<T, U>(x: T, y: T) -> U; - pub(crate) fn simd_lt<T, U>(x: T, y: T) -> U; - pub(crate) fn simd_le<T, U>(x: T, y: T) -> U; - pub(crate) fn simd_gt<T, U>(x: T, y: T) -> U; - pub(crate) fn simd_ge<T, U>(x: T, y: T) -> U; - - pub(crate) fn simd_insert<T, U>(x: T, idx: u32, val: U) -> T; - pub(crate) fn simd_extract<T, U>(x: T, idx: u32) -> U; - - pub(crate) fn simd_cast<T, U>(x: T) -> U; - - pub(crate) fn simd_add<T>(x: T, y: T) -> T; - pub(crate) fn simd_sub<T>(x: T, y: T) -> T; - pub(crate) fn simd_mul<T>(x: T, y: T) -> T; - pub(crate) fn simd_div<T>(x: T, y: T) -> T; - pub(crate) fn simd_rem<T>(x: T, y: T) -> T; - pub(crate) fn simd_shl<T>(x: T, y: T) -> T; - pub(crate) fn simd_shr<T>(x: T, y: T) -> T; - pub(crate) fn simd_and<T>(x: T, y: T) -> T; - pub(crate) fn simd_or<T>(x: T, y: T) -> T; - pub(crate) fn simd_xor<T>(x: T, y: T) -> T; - - pub(crate) fn simd_reduce_add_unordered<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_mul_unordered<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_add_ordered<T, U>(x: T, acc: U) -> U; - pub(crate) fn simd_reduce_mul_ordered<T, U>(x: T, acc: U) -> U; - pub(crate) fn simd_reduce_min<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_max<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_min_nanless<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_max_nanless<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_and<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_or<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_xor<T, U>(x: T) -> U; - pub(crate) fn simd_reduce_all<T>(x: T) -> bool; - pub(crate) fn simd_reduce_any<T>(x: T) -> bool; - - pub(crate) fn simd_select<M, T>(m: M, a: T, b: T) -> T; - - pub(crate) fn simd_fmin<T>(a: T, b: T) -> T; - pub(crate) fn simd_fmax<T>(a: T, b: T) -> T; - - pub(crate) fn simd_fsqrt<T>(a: T) -> T; - pub(crate) fn simd_fma<T>(a: T, b: T, c: T) -> T; - - pub(crate) fn simd_gather<T, P, M>(value: T, pointers: P, mask: M) -> T; - pub(crate) fn simd_scatter<T, P, M>(value: T, pointers: P, mask: M); - - pub(crate) fn simd_bitmask<T, U>(value: T) -> U; -} diff --git a/third_party/rust/packed_simd/src/codegen/math.rs b/third_party/rust/packed_simd/src/codegen/math.rs deleted file mode 100644 index 9a0ea7a4e2..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math.rs +++ /dev/null @@ -1,3 +0,0 @@ -//! Vertical math operations - -pub(crate) mod float; diff --git a/third_party/rust/packed_simd/src/codegen/math/float.rs b/third_party/rust/packed_simd/src/codegen/math/float.rs deleted file mode 100644 index 10d21831f1..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float.rs +++ /dev/null @@ -1,19 +0,0 @@ -//! Vertical floating-point math operations. -#![allow(clippy::useless_transmute)] - -#[macro_use] -pub(crate) mod macros; -pub(crate) mod abs; -pub(crate) mod cos; -pub(crate) mod cos_pi; -pub(crate) mod exp; -pub(crate) mod ln; -pub(crate) mod mul_add; -pub(crate) mod mul_adde; -pub(crate) mod powf; -pub(crate) mod sin; -pub(crate) mod sin_cos_pi; -pub(crate) mod sin_pi; -pub(crate) mod sqrt; -pub(crate) mod sqrte; -pub(crate) mod tanh; diff --git a/third_party/rust/packed_simd/src/codegen/math/float/abs.rs b/third_party/rust/packed_simd/src/codegen/math/float/abs.rs deleted file mode 100644 index 34aacc25be..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/abs.rs +++ /dev/null @@ -1,103 +0,0 @@ -//! Vertical floating-point `fabs` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors fabs - -use crate::*; - -pub(crate) trait Abs { - fn abs(self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.fabs.v2f32"] - fn fabs_v2f32(x: f32x2) -> f32x2; - #[link_name = "llvm.fabs.v4f32"] - fn fabs_v4f32(x: f32x4) -> f32x4; - #[link_name = "llvm.fabs.v8f32"] - fn fabs_v8f32(x: f32x8) -> f32x8; - #[link_name = "llvm.fabs.v16f32"] - fn fabs_v16f32(x: f32x16) -> f32x16; - /* FIXME 64-bit fabsgle elem vectors - #[link_name = "llvm.fabs.v1f64"] - fn fabs_v1f64(x: f64x1) -> f64x1; - */ - #[link_name = "llvm.fabs.v2f64"] - fn fabs_v2f64(x: f64x2) -> f64x2; - #[link_name = "llvm.fabs.v4f64"] - fn fabs_v4f64(x: f64x4) -> f64x4; - #[link_name = "llvm.fabs.v8f64"] - fn fabs_v8f64(x: f64x8) -> f64x8; - - #[link_name = "llvm.fabs.f32"] - fn fabs_f32(x: f32) -> f32; - #[link_name = "llvm.fabs.f64"] - fn fabs_f64(x: f64) -> f64; -} - -gen_unary_impl_table!(Abs, abs); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: fabs_f32); - impl_unary!(f32x4[f32; 4]: fabs_f32); - impl_unary!(f32x8[f32; 8]: fabs_f32); - impl_unary!(f32x16[f32; 16]: fabs_f32); - - impl_unary!(f64x2[f64; 2]: fabs_f64); - impl_unary!(f64x4[f64; 4]: fabs_f64); - impl_unary!(f64x8[f64; 8]: fabs_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_fabsf4_avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_fabsf8_avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_fabsd4_avx2); - - impl_unary!(f32x4: Sleef_fabsf4_avx2128); - impl_unary!(f32x8: Sleef_fabsf8_avx2); - impl_unary!(f64x2: Sleef_fabsd2_avx2128); - impl_unary!(f64x4: Sleef_fabsd4_avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_fabsf4_sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_fabsf8_avx); - impl_unary!(f64x8[h => f64x4]: Sleef_fabsd4_avx); - - impl_unary!(f32x4: Sleef_fabsf4_sse4); - impl_unary!(f32x8: Sleef_fabsf8_avx); - impl_unary!(f64x2: Sleef_fabsd2_sse4); - impl_unary!(f64x4: Sleef_fabsd4_avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_fabsf4_sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_fabsf4_sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_fabsd2_sse4); - - impl_unary!(f32x4: Sleef_fabsf4_sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_fabsf4_sse4); - impl_unary!(f64x2: Sleef_fabsd2_sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_fabsd2_sse4); - } else { - impl_unary!(f32x2[f32; 2]: fabs_f32); - impl_unary!(f32x16: fabs_v16f32); - impl_unary!(f64x8: fabs_v8f64); - - impl_unary!(f32x4: fabs_v4f32); - impl_unary!(f32x8: fabs_v8f32); - impl_unary!(f64x2: fabs_v2f64); - impl_unary!(f64x4: fabs_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: fabs_f32); - impl_unary!(f32x4: fabs_v4f32); - impl_unary!(f32x8: fabs_v8f32); - impl_unary!(f32x16: fabs_v16f32); - - impl_unary!(f64x2: fabs_v2f64); - impl_unary!(f64x4: fabs_v4f64); - impl_unary!(f64x8: fabs_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/cos.rs b/third_party/rust/packed_simd/src/codegen/math/float/cos.rs deleted file mode 100644 index dec390cb74..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/cos.rs +++ /dev/null @@ -1,103 +0,0 @@ -//! Vertical floating-point `cos` -#![allow(unused)] - -// FIXME 64-bit 1 elem vector cos - -use crate::*; - -pub(crate) trait Cos { - fn cos(self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.cos.v2f32"] - fn cos_v2f32(x: f32x2) -> f32x2; - #[link_name = "llvm.cos.v4f32"] - fn cos_v4f32(x: f32x4) -> f32x4; - #[link_name = "llvm.cos.v8f32"] - fn cos_v8f32(x: f32x8) -> f32x8; - #[link_name = "llvm.cos.v16f32"] - fn cos_v16f32(x: f32x16) -> f32x16; - /* FIXME 64-bit cosgle elem vectors - #[link_name = "llvm.cos.v1f64"] - fn cos_v1f64(x: f64x1) -> f64x1; - */ - #[link_name = "llvm.cos.v2f64"] - fn cos_v2f64(x: f64x2) -> f64x2; - #[link_name = "llvm.cos.v4f64"] - fn cos_v4f64(x: f64x4) -> f64x4; - #[link_name = "llvm.cos.v8f64"] - fn cos_v8f64(x: f64x8) -> f64x8; - - #[link_name = "llvm.cos.f32"] - fn cos_f32(x: f32) -> f32; - #[link_name = "llvm.cos.f64"] - fn cos_f64(x: f64) -> f64; -} - -gen_unary_impl_table!(Cos, cos); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: cos_f32); - impl_unary!(f32x4[f32; 4]: cos_f32); - impl_unary!(f32x8[f32; 8]: cos_f32); - impl_unary!(f32x16[f32; 16]: cos_f32); - - impl_unary!(f64x2[f64; 2]: cos_f64); - impl_unary!(f64x4[f64; 4]: cos_f64); - impl_unary!(f64x8[f64; 8]: cos_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_cosf4_u10avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_cosf8_u10avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_cosd4_u10avx2); - - impl_unary!(f32x4: Sleef_cosf4_u10avx2128); - impl_unary!(f32x8: Sleef_cosf8_u10avx2); - impl_unary!(f64x2: Sleef_cosd2_u10avx2128); - impl_unary!(f64x4: Sleef_cosd4_u10avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_cosf4_u10sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_cosf8_u10avx); - impl_unary!(f64x8[h => f64x4]: Sleef_cosd4_u10avx); - - impl_unary!(f32x4: Sleef_cosf4_u10sse4); - impl_unary!(f32x8: Sleef_cosf8_u10avx); - impl_unary!(f64x2: Sleef_cosd2_u10sse4); - impl_unary!(f64x4: Sleef_cosd4_u10avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_cosf4_u10sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_cosf4_u10sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_cosd2_u10sse4); - - impl_unary!(f32x4: Sleef_cosf4_u10sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_cosf4_u10sse4); - impl_unary!(f64x2: Sleef_cosd2_u10sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_cosd2_u10sse4); - } else { - impl_unary!(f32x2[f32; 2]: cos_f32); - impl_unary!(f32x16: cos_v16f32); - impl_unary!(f64x8: cos_v8f64); - - impl_unary!(f32x4: cos_v4f32); - impl_unary!(f32x8: cos_v8f32); - impl_unary!(f64x2: cos_v2f64); - impl_unary!(f64x4: cos_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: cos_f32); - impl_unary!(f32x4: cos_v4f32); - impl_unary!(f32x8: cos_v8f32); - impl_unary!(f32x16: cos_v16f32); - - impl_unary!(f64x2: cos_v2f64); - impl_unary!(f64x4: cos_v4f64); - impl_unary!(f64x8: cos_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs b/third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs deleted file mode 100644 index e283280ee4..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs +++ /dev/null @@ -1,87 +0,0 @@ -//! Vertical floating-point `cos` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors cos_pi - -use crate::*; - -pub(crate) trait CosPi { - fn cos_pi(self) -> Self; -} - -gen_unary_impl_table!(CosPi, cos_pi); - -macro_rules! impl_def { - ($vid:ident, $PI:path) => { - impl CosPi for $vid { - #[inline] - fn cos_pi(self) -> Self { - (self * Self::splat($PI)).cos() - } - } - }; -} -macro_rules! impl_def32 { - ($vid:ident) => { - impl_def!($vid, crate::f32::consts::PI); - }; -} -macro_rules! impl_def64 { - ($vid:ident) => { - impl_def!($vid, crate::f64::consts::PI); - }; -} - -cfg_if! { - if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_cospif4_u05avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_cospif8_u05avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_cospid4_u05avx2); - - impl_unary!(f32x4: Sleef_cospif4_u05avx2128); - impl_unary!(f32x8: Sleef_cospif8_u05avx2); - impl_unary!(f64x2: Sleef_cospid2_u05avx2128); - impl_unary!(f64x4: Sleef_cospid4_u05avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_cospif4_u05sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_cospif8_u05avx); - impl_unary!(f64x8[h => f64x4]: Sleef_cospid4_u05avx); - - impl_unary!(f32x4: Sleef_cospif4_u05sse4); - impl_unary!(f32x8: Sleef_cospif8_u05avx); - impl_unary!(f64x2: Sleef_cospid2_u05sse4); - impl_unary!(f64x4: Sleef_cospid4_u05avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_cospif4_u05sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_cospif4_u05sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_cospid2_u05sse4); - - impl_unary!(f32x4: Sleef_cospif4_u05sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_cospif4_u05sse4); - impl_unary!(f64x2: Sleef_cospid2_u05sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_cospid2_u05sse4); - } else { - impl_def32!(f32x2); - impl_def32!(f32x4); - impl_def32!(f32x8); - impl_def32!(f32x16); - - impl_def64!(f64x2); - impl_def64!(f64x4); - impl_def64!(f64x8); - } - } - } else { - impl_def32!(f32x2); - impl_def32!(f32x4); - impl_def32!(f32x8); - impl_def32!(f32x16); - - impl_def64!(f64x2); - impl_def64!(f64x4); - impl_def64!(f64x8); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/exp.rs b/third_party/rust/packed_simd/src/codegen/math/float/exp.rs deleted file mode 100644 index a7b20580e3..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/exp.rs +++ /dev/null @@ -1,112 +0,0 @@ -//! Vertical floating-point `exp` -#![allow(unused)] - -// FIXME 64-bit expgle elem vectors misexpg - -use crate::*; - -pub(crate) trait Exp { - fn exp(self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.exp.v2f32"] - fn exp_v2f32(x: f32x2) -> f32x2; - #[link_name = "llvm.exp.v4f32"] - fn exp_v4f32(x: f32x4) -> f32x4; - #[link_name = "llvm.exp.v8f32"] - fn exp_v8f32(x: f32x8) -> f32x8; - #[link_name = "llvm.exp.v16f32"] - fn exp_v16f32(x: f32x16) -> f32x16; - /* FIXME 64-bit expgle elem vectors - #[link_name = "llvm.exp.v1f64"] - fn exp_v1f64(x: f64x1) -> f64x1; - */ - #[link_name = "llvm.exp.v2f64"] - fn exp_v2f64(x: f64x2) -> f64x2; - #[link_name = "llvm.exp.v4f64"] - fn exp_v4f64(x: f64x4) -> f64x4; - #[link_name = "llvm.exp.v8f64"] - fn exp_v8f64(x: f64x8) -> f64x8; - - #[link_name = "llvm.exp.f32"] - fn exp_f32(x: f32) -> f32; - #[link_name = "llvm.exp.f64"] - fn exp_f64(x: f64) -> f64; -} - -gen_unary_impl_table!(Exp, exp); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: exp_f32); - impl_unary!(f32x4[f32; 4]: exp_f32); - impl_unary!(f32x8[f32; 8]: exp_f32); - impl_unary!(f32x16[f32; 16]: exp_f32); - - impl_unary!(f64x2[f64; 2]: exp_f64); - impl_unary!(f64x4[f64; 4]: exp_f64); - impl_unary!(f64x8[f64; 8]: exp_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_expf8_u10avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_expd4_u10avx2); - - impl_unary!(f32x4: Sleef_expf4_u10avx2128); - impl_unary!(f32x8: Sleef_expf8_u10avx2); - impl_unary!(f64x2: Sleef_expd2_u10avx2128); - impl_unary!(f64x4: Sleef_expd4_u10avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_expf8_u10avx); - impl_unary!(f64x8[h => f64x4]: Sleef_expd4_u10avx); - - impl_unary!(f32x4: Sleef_expf4_u10sse4); - impl_unary!(f32x8: Sleef_expf8_u10avx); - impl_unary!(f64x2: Sleef_expd2_u10sse4); - impl_unary!(f64x4: Sleef_expd4_u10avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_expf4_u10sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_expd2_u10sse4); - - impl_unary!(f32x4: Sleef_expf4_u10sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_expf4_u10sse4); - impl_unary!(f64x2: Sleef_expd2_u10sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_expd2_u10sse4); - } else if #[cfg(target_feature = "sse2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10sse2); - impl_unary!(f32x16[q => f32x4]: Sleef_expf4_u10sse2); - impl_unary!(f64x8[q => f64x2]: Sleef_expd2_u10sse2); - - impl_unary!(f32x4: Sleef_expf4_u10sse2); - impl_unary!(f32x8[h => f32x4]: Sleef_expf4_u10sse2); - impl_unary!(f64x2: Sleef_expd2_u10sse2); - impl_unary!(f64x4[h => f64x2]: Sleef_expd2_u10sse2); - } else { - impl_unary!(f32x2[f32; 2]: exp_f32); - impl_unary!(f32x16: exp_v16f32); - impl_unary!(f64x8: exp_v8f64); - - impl_unary!(f32x4: exp_v4f32); - impl_unary!(f32x8: exp_v8f32); - impl_unary!(f64x2: exp_v2f64); - impl_unary!(f64x4: exp_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: exp_f32); - impl_unary!(f32x4: exp_v4f32); - impl_unary!(f32x8: exp_v8f32); - impl_unary!(f32x16: exp_v16f32); - - impl_unary!(f64x2: exp_v2f64); - impl_unary!(f64x4: exp_v4f64); - impl_unary!(f64x8: exp_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/ln.rs b/third_party/rust/packed_simd/src/codegen/math/float/ln.rs deleted file mode 100644 index a5e38cb40d..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/ln.rs +++ /dev/null @@ -1,112 +0,0 @@ -//! Vertical floating-point `ln` -#![allow(unused)] - -// FIXME 64-bit lngle elem vectors mislng - -use crate::*; - -pub(crate) trait Ln { - fn ln(self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.log.v2f32"] - fn ln_v2f32(x: f32x2) -> f32x2; - #[link_name = "llvm.log.v4f32"] - fn ln_v4f32(x: f32x4) -> f32x4; - #[link_name = "llvm.log.v8f32"] - fn ln_v8f32(x: f32x8) -> f32x8; - #[link_name = "llvm.log.v16f32"] - fn ln_v16f32(x: f32x16) -> f32x16; - /* FIXME 64-bit lngle elem vectors - #[link_name = "llvm.log.v1f64"] - fn ln_v1f64(x: f64x1) -> f64x1; - */ - #[link_name = "llvm.log.v2f64"] - fn ln_v2f64(x: f64x2) -> f64x2; - #[link_name = "llvm.log.v4f64"] - fn ln_v4f64(x: f64x4) -> f64x4; - #[link_name = "llvm.log.v8f64"] - fn ln_v8f64(x: f64x8) -> f64x8; - - #[link_name = "llvm.log.f32"] - fn ln_f32(x: f32) -> f32; - #[link_name = "llvm.log.f64"] - fn ln_f64(x: f64) -> f64; -} - -gen_unary_impl_table!(Ln, ln); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: ln_f32); - impl_unary!(f32x4[f32; 4]: ln_f32); - impl_unary!(f32x8[f32; 8]: ln_f32); - impl_unary!(f32x16[f32; 16]: ln_f32); - - impl_unary!(f64x2[f64; 2]: ln_f64); - impl_unary!(f64x4[f64; 4]: ln_f64); - impl_unary!(f64x8[f64; 8]: ln_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_logf8_u10avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_logd4_u10avx2); - - impl_unary!(f32x4: Sleef_logf4_u10avx2128); - impl_unary!(f32x8: Sleef_logf8_u10avx2); - impl_unary!(f64x2: Sleef_logd2_u10avx2128); - impl_unary!(f64x4: Sleef_logd4_u10avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_logf8_u10avx); - impl_unary!(f64x8[h => f64x4]: Sleef_logd4_u10avx); - - impl_unary!(f32x4: Sleef_logf4_u10sse4); - impl_unary!(f32x8: Sleef_logf8_u10avx); - impl_unary!(f64x2: Sleef_logd2_u10sse4); - impl_unary!(f64x4: Sleef_logd4_u10avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_logf4_u10sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_logd2_u10sse4); - - impl_unary!(f32x4: Sleef_logf4_u10sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_logf4_u10sse4); - impl_unary!(f64x2: Sleef_logd2_u10sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_logd2_u10sse4); - } else if #[cfg(target_feature = "sse2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10sse2); - impl_unary!(f32x16[q => f32x4]: Sleef_logf4_u10sse2); - impl_unary!(f64x8[q => f64x2]: Sleef_logd2_u10sse2); - - impl_unary!(f32x4: Sleef_logf4_u10sse2); - impl_unary!(f32x8[h => f32x4]: Sleef_logf4_u10sse2); - impl_unary!(f64x2: Sleef_logd2_u10sse2); - impl_unary!(f64x4[h => f64x2]: Sleef_logd2_u10sse2); - } else { - impl_unary!(f32x2[f32; 2]: ln_f32); - impl_unary!(f32x16: ln_v16f32); - impl_unary!(f64x8: ln_v8f64); - - impl_unary!(f32x4: ln_v4f32); - impl_unary!(f32x8: ln_v8f32); - impl_unary!(f64x2: ln_v2f64); - impl_unary!(f64x4: ln_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: ln_f32); - impl_unary!(f32x4: ln_v4f32); - impl_unary!(f32x8: ln_v8f32); - impl_unary!(f32x16: ln_v16f32); - - impl_unary!(f64x2: ln_v2f64); - impl_unary!(f64x4: ln_v4f64); - impl_unary!(f64x8: ln_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/macros.rs b/third_party/rust/packed_simd/src/codegen/math/float/macros.rs deleted file mode 100644 index 8daee1afe2..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/macros.rs +++ /dev/null @@ -1,470 +0,0 @@ -//! Utility macros -#![allow(unused)] - -macro_rules! impl_unary_ { - // implementation mapping 1:1 - (vec | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self) -> Self { - unsafe { - use crate::mem::transmute; - transmute($fun(transmute(self))) - } - } - } - }; - // implementation mapping 1:1 for when `$fun` is a generic function - // like some of the fp math rustc intrinsics (e.g. `fn fun<T>(x: T) -> T`). - (gen | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self) -> Self { - unsafe { - use crate::mem::transmute; - transmute($fun(self.0)) - } - } - } - }; - (scalar | $trait_id:ident, $trait_method:ident, - $vec_id:ident, [$sid:ident; $scount:expr], $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self) -> Self { - unsafe { - union U { - vec: $vec_id, - scalars: [$sid; $scount], - } - let mut scalars = U { vec: self }.scalars; - for i in &mut scalars { - *i = $fun(*i); - } - U { scalars }.vec - } - } - } - }; - // implementation calling fun twice on each of the vector halves: - (halves | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vech_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self) -> Self { - unsafe { - use crate::mem::transmute; - union U { - vec: $vec_id, - halves: [$vech_id; 2], - } - - let mut halves = U { vec: self }.halves; - - *halves.get_unchecked_mut(0) = transmute($fun(transmute(*halves.get_unchecked(0)))); - *halves.get_unchecked_mut(1) = transmute($fun(transmute(*halves.get_unchecked(1)))); - - U { halves }.vec - } - } - } - }; - // implementation calling fun four times on each of the vector quarters: - (quarter | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vecq_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self) -> Self { - unsafe { - use crate::mem::transmute; - union U { - vec: $vec_id, - quarters: [$vecq_id; 4], - } - - let mut quarters = U { vec: self }.quarters; - - *quarters.get_unchecked_mut(0) = transmute($fun(transmute(*quarters.get_unchecked(0)))); - *quarters.get_unchecked_mut(1) = transmute($fun(transmute(*quarters.get_unchecked(1)))); - *quarters.get_unchecked_mut(2) = transmute($fun(transmute(*quarters.get_unchecked(2)))); - *quarters.get_unchecked_mut(3) = transmute($fun(transmute(*quarters.get_unchecked(3)))); - - U { quarters }.vec - } - } - } - }; - // implementation calling fun once on a vector twice as large: - (twice | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vect_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self) -> Self { - unsafe { - use crate::mem::{transmute, uninitialized}; - - union U { - vec: [$vec_id; 2], - twice: $vect_id, - } - - let twice = U { vec: [self, uninitialized()] }.twice; - let twice = transmute($fun(transmute(twice))); - - *(U { twice }.vec.get_unchecked(0)) - } - } - } - }; -} - -macro_rules! gen_unary_impl_table { - ($trait_id:ident, $trait_method:ident) => { - macro_rules! impl_unary { - ($vid:ident: $fun:ident) => { - impl_unary_!(vec | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[g]: $fun:ident) => { - impl_unary_!(gen | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[$sid:ident; $sc:expr]: $fun:ident) => { - impl_unary_!(scalar | $trait_id, $trait_method, $vid, [$sid; $sc], $fun); - }; - ($vid:ident[s]: $fun:ident) => { - impl_unary_!(scalar | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[h => $vid_h:ident]: $fun:ident) => { - impl_unary_!(halves | $trait_id, $trait_method, $vid, $vid_h, $fun); - }; - ($vid:ident[q => $vid_q:ident]: $fun:ident) => { - impl_unary_!(quarter | $trait_id, $trait_method, $vid, $vid_q, $fun); - }; - ($vid:ident[t => $vid_t:ident]: $fun:ident) => { - impl_unary_!(twice | $trait_id, $trait_method, $vid, $vid_t, $fun); - }; - } - }; -} - -macro_rules! impl_tertiary_ { - // implementation mapping 1:1 - (vec | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self, z: Self) -> Self { - unsafe { - use crate::mem::transmute; - transmute($fun(transmute(self), transmute(y), transmute(z))) - } - } - } - }; - (scalar | $trait_id:ident, $trait_method:ident, - $vec_id:ident, [$sid:ident; $scount:expr], $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self, z: Self) -> Self { - unsafe { - union U { - vec: $vec_id, - scalars: [$sid; $scount], - } - let mut x = U { vec: self }.scalars; - let y = U { vec: y }.scalars; - let z = U { vec: z }.scalars; - for (x, (y, z)) in (&mut scalars).zip(&y).zip(&z) { - *i = $fun(*i, *y, *z); - } - U { vec: x }.vec - } - } - } - }; - // implementation calling fun twice on each of the vector halves: - (halves | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vech_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self, z: Self) -> Self { - unsafe { - use crate::mem::transmute; - union U { - vec: $vec_id, - halves: [$vech_id; 2], - } - - let mut x_halves = U { vec: self }.halves; - let y_halves = U { vec: y }.halves; - let z_halves = U { vec: z }.halves; - - *x_halves.get_unchecked_mut(0) = transmute($fun( - transmute(*x_halves.get_unchecked(0)), - transmute(*y_halves.get_unchecked(0)), - transmute(*z_halves.get_unchecked(0)), - )); - *x_halves.get_unchecked_mut(1) = transmute($fun( - transmute(*x_halves.get_unchecked(1)), - transmute(*y_halves.get_unchecked(1)), - transmute(*z_halves.get_unchecked(1)), - )); - - U { halves: x_halves }.vec - } - } - } - }; - // implementation calling fun four times on each of the vector quarters: - (quarter | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vecq_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self, z: Self) -> Self { - unsafe { - use crate::mem::transmute; - union U { - vec: $vec_id, - quarters: [$vecq_id; 4], - } - - let mut x_quarters = U { vec: self }.quarters; - let y_quarters = U { vec: y }.quarters; - let z_quarters = U { vec: z }.quarters; - - *x_quarters.get_unchecked_mut(0) = transmute($fun( - transmute(*x_quarters.get_unchecked(0)), - transmute(*y_quarters.get_unchecked(0)), - transmute(*z_quarters.get_unchecked(0)), - )); - - *x_quarters.get_unchecked_mut(1) = transmute($fun( - transmute(*x_quarters.get_unchecked(1)), - transmute(*y_quarters.get_unchecked(1)), - transmute(*z_quarters.get_unchecked(1)), - )); - - *x_quarters.get_unchecked_mut(2) = transmute($fun( - transmute(*x_quarters.get_unchecked(2)), - transmute(*y_quarters.get_unchecked(2)), - transmute(*z_quarters.get_unchecked(2)), - )); - - *x_quarters.get_unchecked_mut(3) = transmute($fun( - transmute(*x_quarters.get_unchecked(3)), - transmute(*y_quarters.get_unchecked(3)), - transmute(*z_quarters.get_unchecked(3)), - )); - - U { quarters: x_quarters }.vec - } - } - } - }; - // implementation calling fun once on a vector twice as large: - (twice | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vect_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self, z: Self) -> Self { - unsafe { - use crate::mem::{transmute, uninitialized}; - - union U { - vec: [$vec_id; 2], - twice: $vect_id, - } - - let x_twice = U { vec: [self, uninitialized()] }.twice; - let y_twice = U { vec: [y, uninitialized()] }.twice; - let z_twice = U { vec: [z, uninitialized()] }.twice; - let twice: $vect_id = - transmute($fun(transmute(x_twice), transmute(y_twice), transmute(z_twice))); - - *(U { twice }.vec.get_unchecked(0)) - } - } - } - }; -} - -macro_rules! gen_tertiary_impl_table { - ($trait_id:ident, $trait_method:ident) => { - macro_rules! impl_tertiary { - ($vid:ident: $fun:ident) => { - impl_tertiary_!(vec | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[$sid:ident; $sc:expr]: $fun:ident) => { - impl_tertiary_!(scalar | $trait_id, $trait_method, $vid, [$sid; $sc], $fun); - }; - ($vid:ident[s]: $fun:ident) => { - impl_tertiary_!(scalar | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[h => $vid_h:ident]: $fun:ident) => { - impl_tertiary_!(halves | $trait_id, $trait_method, $vid, $vid_h, $fun); - }; - ($vid:ident[q => $vid_q:ident]: $fun:ident) => { - impl_tertiary_!(quarter | $trait_id, $trait_method, $vid, $vid_q, $fun); - }; - ($vid:ident[t => $vid_t:ident]: $fun:ident) => { - impl_tertiary_!(twice | $trait_id, $trait_method, $vid, $vid_t, $fun); - }; - } - }; -} - -macro_rules! impl_binary_ { - // implementation mapping 1:1 - (vec | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self) -> Self { - unsafe { - use crate::mem::transmute; - transmute($fun(transmute(self), transmute(y))) - } - } - } - }; - (scalar | $trait_id:ident, $trait_method:ident, - $vec_id:ident, [$sid:ident; $scount:expr], $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self) -> Self { - unsafe { - union U { - vec: $vec_id, - scalars: [$sid; $scount], - } - let mut x = U { vec: self }.scalars; - let y = U { vec: y }.scalars; - for (x, y) in x.iter_mut().zip(&y) { - *x = $fun(*x, *y); - } - U { scalars: x }.vec - } - } - } - }; - // implementation calling fun twice on each of the vector halves: - (halves | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vech_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self) -> Self { - unsafe { - use crate::mem::transmute; - union U { - vec: $vec_id, - halves: [$vech_id; 2], - } - - let mut x_halves = U { vec: self }.halves; - let y_halves = U { vec: y }.halves; - - *x_halves.get_unchecked_mut(0) = transmute($fun( - transmute(*x_halves.get_unchecked(0)), - transmute(*y_halves.get_unchecked(0)), - )); - *x_halves.get_unchecked_mut(1) = transmute($fun( - transmute(*x_halves.get_unchecked(1)), - transmute(*y_halves.get_unchecked(1)), - )); - - U { halves: x_halves }.vec - } - } - } - }; - // implementation calling fun four times on each of the vector quarters: - (quarter | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vecq_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self) -> Self { - unsafe { - use crate::mem::transmute; - union U { - vec: $vec_id, - quarters: [$vecq_id; 4], - } - - let mut x_quarters = U { vec: self }.quarters; - let y_quarters = U { vec: y }.quarters; - - *x_quarters.get_unchecked_mut(0) = transmute($fun( - transmute(*x_quarters.get_unchecked(0)), - transmute(*y_quarters.get_unchecked(0)), - )); - - *x_quarters.get_unchecked_mut(1) = transmute($fun( - transmute(*x_quarters.get_unchecked(1)), - transmute(*y_quarters.get_unchecked(1)), - )); - - *x_quarters.get_unchecked_mut(2) = transmute($fun( - transmute(*x_quarters.get_unchecked(2)), - transmute(*y_quarters.get_unchecked(2)), - )); - - *x_quarters.get_unchecked_mut(3) = transmute($fun( - transmute(*x_quarters.get_unchecked(3)), - transmute(*y_quarters.get_unchecked(3)), - )); - - U { quarters: x_quarters }.vec - } - } - } - }; - // implementation calling fun once on a vector twice as large: - (twice | $trait_id:ident, $trait_method:ident, $vec_id:ident, - $vect_id:ident, $fun:ident) => { - impl $trait_id for $vec_id { - #[inline] - fn $trait_method(self, y: Self) -> Self { - unsafe { - use crate::mem::{transmute, uninitialized}; - - union U { - vec: [$vec_id; 2], - twice: $vect_id, - } - - let x_twice = U { vec: [self, uninitialized()] }.twice; - let y_twice = U { vec: [y, uninitialized()] }.twice; - let twice: $vect_id = transmute($fun(transmute(x_twice), transmute(y_twice))); - - *(U { twice }.vec.get_unchecked(0)) - } - } - } - }; -} - -macro_rules! gen_binary_impl_table { - ($trait_id:ident, $trait_method:ident) => { - macro_rules! impl_binary { - ($vid:ident: $fun:ident) => { - impl_binary_!(vec | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[$sid:ident; $sc:expr]: $fun:ident) => { - impl_binary_!(scalar | $trait_id, $trait_method, $vid, [$sid; $sc], $fun); - }; - ($vid:ident[s]: $fun:ident) => { - impl_binary_!(scalar | $trait_id, $trait_method, $vid, $fun); - }; - ($vid:ident[h => $vid_h:ident]: $fun:ident) => { - impl_binary_!(halves | $trait_id, $trait_method, $vid, $vid_h, $fun); - }; - ($vid:ident[q => $vid_q:ident]: $fun:ident) => { - impl_binary_!(quarter | $trait_id, $trait_method, $vid, $vid_q, $fun); - }; - ($vid:ident[t => $vid_t:ident]: $fun:ident) => { - impl_binary_!(twice | $trait_id, $trait_method, $vid, $vid_t, $fun); - }; - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs b/third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs deleted file mode 100644 index d37f30fa86..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs +++ /dev/null @@ -1,109 +0,0 @@ -//! Vertical floating-point `mul_add` -#![allow(unused)] -use crate::*; - -// FIXME: 64-bit 1 element mul_add - -pub(crate) trait MulAdd { - fn mul_add(self, y: Self, z: Self) -> Self; -} - -#[cfg(not(target_arch = "s390x"))] -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.fma.v2f32"] - fn fma_v2f32(x: f32x2, y: f32x2, z: f32x2) -> f32x2; - #[link_name = "llvm.fma.v4f32"] - fn fma_v4f32(x: f32x4, y: f32x4, z: f32x4) -> f32x4; - #[link_name = "llvm.fma.v8f32"] - fn fma_v8f32(x: f32x8, y: f32x8, z: f32x8) -> f32x8; - #[link_name = "llvm.fma.v16f32"] - fn fma_v16f32(x: f32x16, y: f32x16, z: f32x16) -> f32x16; - /* FIXME 64-bit single elem vectors - #[link_name = "llvm.fma.v1f64"] - fn fma_v1f64(x: f64x1, y: f64x1, z: f64x1) -> f64x1; - */ - #[link_name = "llvm.fma.v2f64"] - fn fma_v2f64(x: f64x2, y: f64x2, z: f64x2) -> f64x2; - #[link_name = "llvm.fma.v4f64"] - fn fma_v4f64(x: f64x4, y: f64x4, z: f64x4) -> f64x4; - #[link_name = "llvm.fma.v8f64"] - fn fma_v8f64(x: f64x8, y: f64x8, z: f64x8) -> f64x8; -} - -gen_tertiary_impl_table!(MulAdd, mul_add); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - macro_rules! impl_broken { - ($id:ident) => { - impl MulAdd for $id { - #[inline] - fn mul_add(self, y: Self, z: Self) -> Self { - self * y + z - } - } - }; - } - - impl_broken!(f32x2); - impl_broken!(f32x4); - impl_broken!(f32x8); - impl_broken!(f32x16); - - impl_broken!(f64x2); - impl_broken!(f64x4); - impl_broken!(f64x8); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_tertiary!(f32x2[t => f32x4]: Sleef_fmaf4_avx2128); - impl_tertiary!(f32x16[h => f32x8]: Sleef_fmaf8_avx2); - impl_tertiary!(f64x8[h => f64x4]: Sleef_fmad4_avx2); - - impl_tertiary!(f32x4: Sleef_fmaf4_avx2128); - impl_tertiary!(f32x8: Sleef_fmaf8_avx2); - impl_tertiary!(f64x2: Sleef_fmad2_avx2128); - impl_tertiary!(f64x4: Sleef_fmad4_avx2); - } else if #[cfg(target_feature = "avx")] { - impl_tertiary!(f32x2[t => f32x4]: Sleef_fmaf4_sse4); - impl_tertiary!(f32x16[h => f32x8]: Sleef_fmaf8_avx); - impl_tertiary!(f64x8[h => f64x4]: Sleef_fmad4_avx); - - impl_tertiary!(f32x4: Sleef_fmaf4_sse4); - impl_tertiary!(f32x8: Sleef_fmaf8_avx); - impl_tertiary!(f64x2: Sleef_fmad2_sse4); - impl_tertiary!(f64x4: Sleef_fmad4_avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_tertiary!(f32x2[t => f32x4]: Sleef_fmaf4_sse4); - impl_tertiary!(f32x16[q => f32x4]: Sleef_fmaf4_sse4); - impl_tertiary!(f64x8[q => f64x2]: Sleef_fmad2_sse4); - - impl_tertiary!(f32x4: Sleef_fmaf4_sse4); - impl_tertiary!(f32x8[h => f32x4]: Sleef_fmaf4_sse4); - impl_tertiary!(f64x2: Sleef_fmad2_sse4); - impl_tertiary!(f64x4[h => f64x2]: Sleef_fmad2_sse4); - } else { - impl_tertiary!(f32x2: fma_v2f32); - impl_tertiary!(f32x16: fma_v16f32); - impl_tertiary!(f64x8: fma_v8f64); - - impl_tertiary!(f32x4: fma_v4f32); - impl_tertiary!(f32x8: fma_v8f32); - impl_tertiary!(f64x2: fma_v2f64); - impl_tertiary!(f64x4: fma_v4f64); - } - } - } else { - impl_tertiary!(f32x2: fma_v2f32); - impl_tertiary!(f32x4: fma_v4f32); - impl_tertiary!(f32x8: fma_v8f32); - impl_tertiary!(f32x16: fma_v16f32); - // impl_tertiary!(f64x1: fma_v1f64); // FIXME 64-bit fmagle elem vectors - impl_tertiary!(f64x2: fma_v2f64); - impl_tertiary!(f64x4: fma_v4f64); - impl_tertiary!(f64x8: fma_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs b/third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs deleted file mode 100644 index c0baeacec2..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs +++ /dev/null @@ -1,60 +0,0 @@ -//! Approximation for floating-point `mul_add` -use crate::*; - -// FIXME: 64-bit 1 element mul_adde - -pub(crate) trait MulAddE { - fn mul_adde(self, y: Self, z: Self) -> Self; -} - -#[cfg(not(target_arch = "s390x"))] -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.fmuladd.v2f32"] - fn fmuladd_v2f32(x: f32x2, y: f32x2, z: f32x2) -> f32x2; - #[link_name = "llvm.fmuladd.v4f32"] - fn fmuladd_v4f32(x: f32x4, y: f32x4, z: f32x4) -> f32x4; - #[link_name = "llvm.fmuladd.v8f32"] - fn fmuladd_v8f32(x: f32x8, y: f32x8, z: f32x8) -> f32x8; - #[link_name = "llvm.fmuladd.v16f32"] - fn fmuladd_v16f32(x: f32x16, y: f32x16, z: f32x16) -> f32x16; - /* FIXME 64-bit single elem vectors - #[link_name = "llvm.fmuladd.v1f64"] - fn fmuladd_v1f64(x: f64x1, y: f64x1, z: f64x1) -> f64x1; - */ - #[link_name = "llvm.fmuladd.v2f64"] - fn fmuladd_v2f64(x: f64x2, y: f64x2, z: f64x2) -> f64x2; - #[link_name = "llvm.fmuladd.v4f64"] - fn fmuladd_v4f64(x: f64x4, y: f64x4, z: f64x4) -> f64x4; - #[link_name = "llvm.fmuladd.v8f64"] - fn fmuladd_v8f64(x: f64x8, y: f64x8, z: f64x8) -> f64x8; -} - -macro_rules! impl_mul_adde { - ($id:ident : $fn:ident) => { - impl MulAddE for $id { - #[inline] - fn mul_adde(self, y: Self, z: Self) -> Self { - #[cfg(not(target_arch = "s390x"))] - { - use crate::mem::transmute; - unsafe { transmute($fn(transmute(self), transmute(y), transmute(z))) } - } - #[cfg(target_arch = "s390x")] - { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - self * y + z - } - } - } - }; -} - -impl_mul_adde!(f32x2: fmuladd_v2f32); -impl_mul_adde!(f32x4: fmuladd_v4f32); -impl_mul_adde!(f32x8: fmuladd_v8f32); -impl_mul_adde!(f32x16: fmuladd_v16f32); -// impl_mul_adde!(f64x1: fma_v1f64); // FIXME 64-bit fmagle elem vectors -impl_mul_adde!(f64x2: fmuladd_v2f64); -impl_mul_adde!(f64x4: fmuladd_v4f64); -impl_mul_adde!(f64x8: fmuladd_v8f64); diff --git a/third_party/rust/packed_simd/src/codegen/math/float/powf.rs b/third_party/rust/packed_simd/src/codegen/math/float/powf.rs deleted file mode 100644 index 89ca52e96d..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/powf.rs +++ /dev/null @@ -1,112 +0,0 @@ -//! Vertical floating-point `powf` -#![allow(unused)] - -// FIXME 64-bit powfgle elem vectors mispowfg - -use crate::*; - -pub(crate) trait Powf { - fn powf(self, x: Self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.pow.v2f32"] - fn powf_v2f32(x: f32x2, y: f32x2) -> f32x2; - #[link_name = "llvm.pow.v4f32"] - fn powf_v4f32(x: f32x4, y: f32x4) -> f32x4; - #[link_name = "llvm.pow.v8f32"] - fn powf_v8f32(x: f32x8, y: f32x8) -> f32x8; - #[link_name = "llvm.pow.v16f32"] - fn powf_v16f32(x: f32x16, y: f32x16) -> f32x16; - /* FIXME 64-bit powfgle elem vectors - #[link_name = "llvm.pow.v1f64"] - fn powf_v1f64(x: f64x1, y: f64x1) -> f64x1; - */ - #[link_name = "llvm.pow.v2f64"] - fn powf_v2f64(x: f64x2, y: f64x2) -> f64x2; - #[link_name = "llvm.pow.v4f64"] - fn powf_v4f64(x: f64x4, y: f64x4) -> f64x4; - #[link_name = "llvm.pow.v8f64"] - fn powf_v8f64(x: f64x8, y: f64x8) -> f64x8; - - #[link_name = "llvm.pow.f32"] - fn powf_f32(x: f32, y: f32) -> f32; - #[link_name = "llvm.pow.f64"] - fn powf_f64(x: f64, y: f64) -> f64; -} - -gen_binary_impl_table!(Powf, powf); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_binary!(f32x2[f32; 2]: powf_f32); - impl_binary!(f32x4[f32; 4]: powf_f32); - impl_binary!(f32x8[f32; 8]: powf_f32); - impl_binary!(f32x16[f32; 16]: powf_f32); - - impl_binary!(f64x2[f64; 2]: powf_f64); - impl_binary!(f64x4[f64; 4]: powf_f64); - impl_binary!(f64x8[f64; 8]: powf_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10avx2128); - impl_binary!(f32x16[h => f32x8]: Sleef_powf8_u10avx2); - impl_binary!(f64x8[h => f64x4]: Sleef_powd4_u10avx2); - - impl_binary!(f32x4: Sleef_powf4_u10avx2128); - impl_binary!(f32x8: Sleef_powf8_u10avx2); - impl_binary!(f64x2: Sleef_powd2_u10avx2128); - impl_binary!(f64x4: Sleef_powd4_u10avx2); - } else if #[cfg(target_feature = "avx")] { - impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10sse4); - impl_binary!(f32x16[h => f32x8]: Sleef_powf8_u10avx); - impl_binary!(f64x8[h => f64x4]: Sleef_powd4_u10avx); - - impl_binary!(f32x4: Sleef_powf4_u10sse4); - impl_binary!(f32x8: Sleef_powf8_u10avx); - impl_binary!(f64x2: Sleef_powd2_u10sse4); - impl_binary!(f64x4: Sleef_powd4_u10avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10sse4); - impl_binary!(f32x16[q => f32x4]: Sleef_powf4_u10sse4); - impl_binary!(f64x8[q => f64x2]: Sleef_powd2_u10sse4); - - impl_binary!(f32x4: Sleef_powf4_u10sse4); - impl_binary!(f32x8[h => f32x4]: Sleef_powf4_u10sse4); - impl_binary!(f64x2: Sleef_powd2_u10sse4); - impl_binary!(f64x4[h => f64x2]: Sleef_powd2_u10sse4); - } else if #[cfg(target_feature = "sse2")] { - impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10sse2); - impl_binary!(f32x16[q => f32x4]: Sleef_powf4_u10sse2); - impl_binary!(f64x8[q => f64x2]: Sleef_powd2_u10sse2); - - impl_binary!(f32x4: Sleef_powf4_u10sse2); - impl_binary!(f32x8[h => f32x4]: Sleef_powf4_u10sse2); - impl_binary!(f64x2: Sleef_powd2_u10sse2); - impl_binary!(f64x4[h => f64x2]: Sleef_powd2_u10sse2); - } else { - impl_binary!(f32x2[f32; 2]: powf_f32); - impl_binary!(f32x4: powf_v4f32); - impl_binary!(f32x8: powf_v8f32); - impl_binary!(f32x16: powf_v16f32); - - impl_binary!(f64x2: powf_v2f64); - impl_binary!(f64x4: powf_v4f64); - impl_binary!(f64x8: powf_v8f64); - } - } - } else { - impl_binary!(f32x2[f32; 2]: powf_f32); - impl_binary!(f32x4: powf_v4f32); - impl_binary!(f32x8: powf_v8f32); - impl_binary!(f32x16: powf_v16f32); - - impl_binary!(f64x2: powf_v2f64); - impl_binary!(f64x4: powf_v4f64); - impl_binary!(f64x8: powf_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sin.rs b/third_party/rust/packed_simd/src/codegen/math/float/sin.rs deleted file mode 100644 index d881415909..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/sin.rs +++ /dev/null @@ -1,103 +0,0 @@ -//! Vertical floating-point `sin` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors sin - -use crate::*; - -pub(crate) trait Sin { - fn sin(self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.sin.v2f32"] - fn sin_v2f32(x: f32x2) -> f32x2; - #[link_name = "llvm.sin.v4f32"] - fn sin_v4f32(x: f32x4) -> f32x4; - #[link_name = "llvm.sin.v8f32"] - fn sin_v8f32(x: f32x8) -> f32x8; - #[link_name = "llvm.sin.v16f32"] - fn sin_v16f32(x: f32x16) -> f32x16; - /* FIXME 64-bit single elem vectors - #[link_name = "llvm.sin.v1f64"] - fn sin_v1f64(x: f64x1) -> f64x1; - */ - #[link_name = "llvm.sin.v2f64"] - fn sin_v2f64(x: f64x2) -> f64x2; - #[link_name = "llvm.sin.v4f64"] - fn sin_v4f64(x: f64x4) -> f64x4; - #[link_name = "llvm.sin.v8f64"] - fn sin_v8f64(x: f64x8) -> f64x8; - - #[link_name = "llvm.sin.f32"] - fn sin_f32(x: f32) -> f32; - #[link_name = "llvm.sin.f64"] - fn sin_f64(x: f64) -> f64; -} - -gen_unary_impl_table!(Sin, sin); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: sin_f32); - impl_unary!(f32x4[f32; 4]: sin_f32); - impl_unary!(f32x8[f32; 8]: sin_f32); - impl_unary!(f32x16[f32; 16]: sin_f32); - - impl_unary!(f64x2[f64; 2]: sin_f64); - impl_unary!(f64x4[f64; 4]: sin_f64); - impl_unary!(f64x8[f64; 8]: sin_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sinf4_u10avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_sinf8_u10avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_sind4_u10avx2); - - impl_unary!(f32x4: Sleef_sinf4_u10avx2128); - impl_unary!(f32x8: Sleef_sinf8_u10avx2); - impl_unary!(f64x2: Sleef_sind2_u10avx2128); - impl_unary!(f64x4: Sleef_sind4_u10avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sinf4_u10sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_sinf8_u10avx); - impl_unary!(f64x8[h => f64x4]: Sleef_sind4_u10avx); - - impl_unary!(f32x4: Sleef_sinf4_u10sse4); - impl_unary!(f32x8: Sleef_sinf8_u10avx); - impl_unary!(f64x2: Sleef_sind2_u10sse4); - impl_unary!(f64x4: Sleef_sind4_u10avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sinf4_u10sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_sinf4_u10sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_sind2_u10sse4); - - impl_unary!(f32x4: Sleef_sinf4_u10sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_sinf4_u10sse4); - impl_unary!(f64x2: Sleef_sind2_u10sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_sind2_u10sse4); - } else { - impl_unary!(f32x2[f32; 2]: sin_f32); - impl_unary!(f32x16: sin_v16f32); - impl_unary!(f64x8: sin_v8f64); - - impl_unary!(f32x4: sin_v4f32); - impl_unary!(f32x8: sin_v8f32); - impl_unary!(f64x2: sin_v2f64); - impl_unary!(f64x4: sin_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: sin_f32); - impl_unary!(f32x4: sin_v4f32); - impl_unary!(f32x8: sin_v8f32); - impl_unary!(f32x16: sin_v16f32); - - impl_unary!(f64x2: sin_v2f64); - impl_unary!(f64x4: sin_v4f64); - impl_unary!(f64x8: sin_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs b/third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs deleted file mode 100644 index b283d11111..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs +++ /dev/null @@ -1,188 +0,0 @@ -//! Vertical floating-point `sin_cos` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors sin_cos - -use crate::*; - -pub(crate) trait SinCosPi: Sized { - type Output; - fn sin_cos_pi(self) -> Self::Output; -} - -macro_rules! impl_def { - ($vid:ident, $PI:path) => { - impl SinCosPi for $vid { - type Output = (Self, Self); - #[inline] - fn sin_cos_pi(self) -> Self::Output { - let v = self * Self::splat($PI); - (v.sin(), v.cos()) - } - } - }; -} - -macro_rules! impl_def32 { - ($vid:ident) => { - impl_def!($vid, crate::f32::consts::PI); - }; -} -macro_rules! impl_def64 { - ($vid:ident) => { - impl_def!($vid, crate::f64::consts::PI); - }; -} - -macro_rules! impl_unary_t { - ($vid:ident: $fun:ident) => { - impl SinCosPi for $vid { - type Output = (Self, Self); - fn sin_cos_pi(self) -> Self::Output { - unsafe { - use crate::mem::transmute; - transmute($fun(transmute(self))) - } - } - } - }; - ($vid:ident[t => $vid_t:ident]: $fun:ident) => { - impl SinCosPi for $vid { - type Output = (Self, Self); - fn sin_cos_pi(self) -> Self::Output { - unsafe { - use crate::mem::{transmute, uninitialized}; - - union U { - vec: [$vid; 2], - twice: $vid_t, - } - - let twice = U { vec: [self, uninitialized()] }.twice; - let twice = transmute($fun(transmute(twice))); - - union R { - twice: ($vid_t, $vid_t), - vecs: ([$vid; 2], [$vid; 2]), - } - let r = R { twice }.vecs; - (*r.0.get_unchecked(0), *r.0.get_unchecked(1)) - } - } - } - }; - ($vid:ident[h => $vid_h:ident]: $fun:ident) => { - impl SinCosPi for $vid { - type Output = (Self, Self); - fn sin_cos_pi(self) -> Self::Output { - unsafe { - use crate::mem::transmute; - - union U { - vec: $vid, - halves: [$vid_h; 2], - } - - let halves = U { vec: self }.halves; - - let res_0: ($vid_h, $vid_h) = transmute($fun(transmute(*halves.get_unchecked(0)))); - let res_1: ($vid_h, $vid_h) = transmute($fun(transmute(*halves.get_unchecked(1)))); - - union R { - result: ($vid, $vid), - halves: ([$vid_h; 2], [$vid_h; 2]), - } - R { halves: ([res_0.0, res_1.0], [res_0.1, res_1.1]) }.result - } - } - } - }; - ($vid:ident[q => $vid_q:ident]: $fun:ident) => { - impl SinCosPi for $vid { - type Output = (Self, Self); - fn sin_cos_pi(self) -> Self::Output { - unsafe { - use crate::mem::transmute; - - union U { - vec: $vid, - quarters: [$vid_q; 4], - } - - let quarters = U { vec: self }.quarters; - - let res_0: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(0)))); - let res_1: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(1)))); - let res_2: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(2)))); - let res_3: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(3)))); - - union R { - result: ($vid, $vid), - quarters: ([$vid_q; 4], [$vid_q; 4]), - } - R { - quarters: ( - [res_0.0, res_1.0, res_2.0, res_3.0], - [res_0.1, res_1.1, res_2.1, res_3.1], - ), - } - .result - } - } - } - }; -} - -cfg_if! { - if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary_t!(f32x2[t => f32x4]: Sleef_sincospif4_u05avx2128); - impl_unary_t!(f32x16[h => f32x8]: Sleef_sincospif8_u05avx2); - impl_unary_t!(f64x8[h => f64x4]: Sleef_sincospid4_u05avx2); - - impl_unary_t!(f32x4: Sleef_sincospif4_u05avx2128); - impl_unary_t!(f32x8: Sleef_sincospif8_u05avx2); - impl_unary_t!(f64x2: Sleef_sincospid2_u05avx2128); - impl_unary_t!(f64x4: Sleef_sincospid4_u05avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary_t!(f32x2[t => f32x4]: Sleef_sincospif4_u05sse4); - impl_unary_t!(f32x16[h => f32x8]: Sleef_sincospif8_u05avx); - impl_unary_t!(f64x8[h => f64x4]: Sleef_sincospid4_u05avx); - - impl_unary_t!(f32x4: Sleef_sincospif4_u05sse4); - impl_unary_t!(f32x8: Sleef_sincospif8_u05avx); - impl_unary_t!(f64x2: Sleef_sincospid2_u05sse4); - impl_unary_t!(f64x4: Sleef_sincospid4_u05avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary_t!(f32x2[t => f32x4]: Sleef_sincospif4_u05sse4); - impl_unary_t!(f32x16[q => f32x4]: Sleef_sincospif4_u05sse4); - impl_unary_t!(f64x8[q => f64x2]: Sleef_sincospid2_u05sse4); - - impl_unary_t!(f32x4: Sleef_sincospif4_u05sse4); - impl_unary_t!(f32x8[h => f32x4]: Sleef_sincospif4_u05sse4); - impl_unary_t!(f64x2: Sleef_sincospid2_u05sse4); - impl_unary_t!(f64x4[h => f64x2]: Sleef_sincospid2_u05sse4); - } else { - impl_def32!(f32x2); - impl_def32!(f32x4); - impl_def32!(f32x8); - impl_def32!(f32x16); - - impl_def64!(f64x2); - impl_def64!(f64x4); - impl_def64!(f64x8); - } - } - } else { - impl_def32!(f32x2); - impl_def32!(f32x4); - impl_def32!(f32x8); - impl_def32!(f32x16); - - impl_def64!(f64x2); - impl_def64!(f64x4); - impl_def64!(f64x8); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs b/third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs deleted file mode 100644 index 0c8f6bb120..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs +++ /dev/null @@ -1,87 +0,0 @@ -//! Vertical floating-point `sin_pi` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors sin_pi - -use crate::*; - -pub(crate) trait SinPi { - fn sin_pi(self) -> Self; -} - -gen_unary_impl_table!(SinPi, sin_pi); - -macro_rules! impl_def { - ($vid:ident, $PI:path) => { - impl SinPi for $vid { - #[inline] - fn sin_pi(self) -> Self { - (self * Self::splat($PI)).sin() - } - } - }; -} -macro_rules! impl_def32 { - ($vid:ident) => { - impl_def!($vid, crate::f32::consts::PI); - }; -} -macro_rules! impl_def64 { - ($vid:ident) => { - impl_def!($vid, crate::f64::consts::PI); - }; -} - -cfg_if! { - if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sinpif4_u05avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_sinpif8_u05avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_sinpid4_u05avx2); - - impl_unary!(f32x4: Sleef_sinpif4_u05avx2128); - impl_unary!(f32x8: Sleef_sinpif8_u05avx2); - impl_unary!(f64x2: Sleef_sinpid2_u05avx2128); - impl_unary!(f64x4: Sleef_sinpid4_u05avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sinpif4_u05sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_sinpif8_u05avx); - impl_unary!(f64x8[h => f64x4]: Sleef_sinpid4_u05avx); - - impl_unary!(f32x4: Sleef_sinpif4_u05sse4); - impl_unary!(f32x8: Sleef_sinpif8_u05avx); - impl_unary!(f64x2: Sleef_sinpid2_u05sse4); - impl_unary!(f64x4: Sleef_sinpid4_u05avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sinpif4_u05sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_sinpif4_u05sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_sinpid2_u05sse4); - - impl_unary!(f32x4: Sleef_sinpif4_u05sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_sinpif4_u05sse4); - impl_unary!(f64x2: Sleef_sinpid2_u05sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_sinpid2_u05sse4); - } else { - impl_def32!(f32x2); - impl_def32!(f32x4); - impl_def32!(f32x8); - impl_def32!(f32x16); - - impl_def64!(f64x2); - impl_def64!(f64x4); - impl_def64!(f64x8); - } - } - } else { - impl_def32!(f32x2); - impl_def32!(f32x4); - impl_def32!(f32x8); - impl_def32!(f32x16); - - impl_def64!(f64x2); - impl_def64!(f64x4); - impl_def64!(f64x8); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs b/third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs deleted file mode 100644 index 67bb0a2a9c..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs +++ /dev/null @@ -1,103 +0,0 @@ -//! Vertical floating-point `sqrt` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors sqrt - -use crate::*; - -pub(crate) trait Sqrt { - fn sqrt(self) -> Self; -} - -#[allow(improper_ctypes)] -extern "C" { - #[link_name = "llvm.sqrt.v2f32"] - fn sqrt_v2f32(x: f32x2) -> f32x2; - #[link_name = "llvm.sqrt.v4f32"] - fn sqrt_v4f32(x: f32x4) -> f32x4; - #[link_name = "llvm.sqrt.v8f32"] - fn sqrt_v8f32(x: f32x8) -> f32x8; - #[link_name = "llvm.sqrt.v16f32"] - fn sqrt_v16f32(x: f32x16) -> f32x16; - /* FIXME 64-bit sqrtgle elem vectors - #[link_name = "llvm.sqrt.v1f64"] - fn sqrt_v1f64(x: f64x1) -> f64x1; - */ - #[link_name = "llvm.sqrt.v2f64"] - fn sqrt_v2f64(x: f64x2) -> f64x2; - #[link_name = "llvm.sqrt.v4f64"] - fn sqrt_v4f64(x: f64x4) -> f64x4; - #[link_name = "llvm.sqrt.v8f64"] - fn sqrt_v8f64(x: f64x8) -> f64x8; - - #[link_name = "llvm.sqrt.f32"] - fn sqrt_f32(x: f32) -> f32; - #[link_name = "llvm.sqrt.f64"] - fn sqrt_f64(x: f64) -> f64; -} - -gen_unary_impl_table!(Sqrt, sqrt); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: sqrt_f32); - impl_unary!(f32x4[f32; 4]: sqrt_f32); - impl_unary!(f32x8[f32; 8]: sqrt_f32); - impl_unary!(f32x16[f32; 16]: sqrt_f32); - - impl_unary!(f64x2[f64; 2]: sqrt_f64); - impl_unary!(f64x4[f64; 4]: sqrt_f64); - impl_unary!(f64x8[f64; 8]: sqrt_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_avx2); - - impl_unary!(f32x4: Sleef_sqrtf4_avx2128); - impl_unary!(f32x8: Sleef_sqrtf8_avx2); - impl_unary!(f64x2: Sleef_sqrtd2_avx2128); - impl_unary!(f64x4: Sleef_sqrtd4_avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_avx); - impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_avx); - - impl_unary!(f32x4: Sleef_sqrtf4_sse4); - impl_unary!(f32x8: Sleef_sqrtf8_avx); - impl_unary!(f64x2: Sleef_sqrtd2_sse4); - impl_unary!(f64x4: Sleef_sqrtd4_avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_sqrtf4_sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_sqrtd2_sse4); - - impl_unary!(f32x4: Sleef_sqrtf4_sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_sqrtf4_sse4); - impl_unary!(f64x2: Sleef_sqrtd2_sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_sqrtd2_sse4); - } else { - impl_unary!(f32x2[f32; 2]: sqrt_f32); - impl_unary!(f32x16: sqrt_v16f32); - impl_unary!(f64x8: sqrt_v8f64); - - impl_unary!(f32x4: sqrt_v4f32); - impl_unary!(f32x8: sqrt_v8f32); - impl_unary!(f64x2: sqrt_v2f64); - impl_unary!(f64x4: sqrt_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: sqrt_f32); - impl_unary!(f32x4: sqrt_v4f32); - impl_unary!(f32x8: sqrt_v8f32); - impl_unary!(f32x16: sqrt_v16f32); - - impl_unary!(f64x2: sqrt_v2f64); - impl_unary!(f64x4: sqrt_v4f64); - impl_unary!(f64x8: sqrt_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs b/third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs deleted file mode 100644 index 58a1de1f40..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs +++ /dev/null @@ -1,67 +0,0 @@ -//! Vertical floating-point `sqrt` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors sqrte - -use crate::llvm::simd_fsqrt; -use crate::*; - -pub(crate) trait Sqrte { - fn sqrte(self) -> Self; -} - -gen_unary_impl_table!(Sqrte, sqrte); - -cfg_if! { - if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_u35avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_u35avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_u35avx2); - - impl_unary!(f32x4: Sleef_sqrtf4_u35avx2128); - impl_unary!(f32x8: Sleef_sqrtf8_u35avx2); - impl_unary!(f64x2: Sleef_sqrtd2_u35avx2128); - impl_unary!(f64x4: Sleef_sqrtd4_u35avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_u35sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_u35avx); - impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_u35avx); - - impl_unary!(f32x4: Sleef_sqrtf4_u35sse4); - impl_unary!(f32x8: Sleef_sqrtf8_u35avx); - impl_unary!(f64x2: Sleef_sqrtd2_u35sse4); - impl_unary!(f64x4: Sleef_sqrtd4_u35avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_u35sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_sqrtf4_u35sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_sqrtd2_u35sse4); - - impl_unary!(f32x4: Sleef_sqrtf4_u35sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_sqrtf4_u35sse4); - impl_unary!(f64x2: Sleef_sqrtd2_u35sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_sqrtd2_u35sse4); - } else { - impl_unary!(f32x2[g]: simd_fsqrt); - impl_unary!(f32x16[g]: simd_fsqrt); - impl_unary!(f64x8[g]: simd_fsqrt); - - impl_unary!(f32x4[g]: simd_fsqrt); - impl_unary!(f32x8[g]: simd_fsqrt); - impl_unary!(f64x2[g]: simd_fsqrt); - impl_unary!(f64x4[g]: simd_fsqrt); - } - } - } else { - impl_unary!(f32x2[g]: simd_fsqrt); - impl_unary!(f32x4[g]: simd_fsqrt); - impl_unary!(f32x8[g]: simd_fsqrt); - impl_unary!(f32x16[g]: simd_fsqrt); - - impl_unary!(f64x2[g]: simd_fsqrt); - impl_unary!(f64x4[g]: simd_fsqrt); - impl_unary!(f64x8[g]: simd_fsqrt); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/tanh.rs b/third_party/rust/packed_simd/src/codegen/math/float/tanh.rs deleted file mode 100644 index 4243b0d886..0000000000 --- a/third_party/rust/packed_simd/src/codegen/math/float/tanh.rs +++ /dev/null @@ -1,120 +0,0 @@ -//! Vertical floating-point `tanh` -#![allow(unused)] - -// FIXME 64-bit 1 elem vectors tanh - -#[cfg(not(feature = "std"))] -use num_traits::Float; - -use crate::*; - -pub(crate) trait Tanh { - fn tanh(self) -> Self; -} - -macro_rules! define_tanh { - ($name:ident, $basetype:ty, $simdtype:ty, $lanes:expr, $trait:path) => { - fn $name(x: $simdtype) -> $simdtype { - use core::intrinsics::transmute; - let mut buf: [$basetype; $lanes] = unsafe { transmute(x) }; - for elem in &mut buf { - *elem = <$basetype as $trait>::tanh(*elem); - } - unsafe { transmute(buf) } - } - }; - - (f32 => $name:ident, $type:ty, $lanes:expr) => { - define_tanh!($name, f32, $type, $lanes, Float); - }; - - (f64 => $name:ident, $type:ty, $lanes:expr) => { - define_tanh!($name, f64, $type, $lanes, Float); - }; -} - -// llvm does not seem to expose the hyperbolic versions of trigonometric -// functions; we thus call the classical rust versions on all of them (which -// stem from cmath). -define_tanh!(f32 => tanh_v2f32, f32x2, 2); -define_tanh!(f32 => tanh_v4f32, f32x4, 4); -define_tanh!(f32 => tanh_v8f32, f32x8, 8); -define_tanh!(f32 => tanh_v16f32, f32x16, 16); - -define_tanh!(f64 => tanh_v2f64, f64x2, 2); -define_tanh!(f64 => tanh_v4f64, f64x4, 4); -define_tanh!(f64 => tanh_v8f64, f64x8, 8); - -fn tanh_f32(x: f32) -> f32 { - Float::tanh(x) -} - -fn tanh_f64(x: f64) -> f64 { - Float::tanh(x) -} - -gen_unary_impl_table!(Tanh, tanh); - -cfg_if! { - if #[cfg(target_arch = "s390x")] { - // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 - impl_unary!(f32x2[f32; 2]: tanh_f32); - impl_unary!(f32x4[f32; 4]: tanh_f32); - impl_unary!(f32x8[f32; 8]: tanh_f32); - impl_unary!(f32x16[f32; 16]: tanh_f32); - - impl_unary!(f64x2[f64; 2]: tanh_f64); - impl_unary!(f64x4[f64; 4]: tanh_f64); - impl_unary!(f64x8[f64; 8]: tanh_f64); - } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { - use sleef_sys::*; - cfg_if! { - if #[cfg(target_feature = "avx2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_tanhf4_u10avx2128); - impl_unary!(f32x16[h => f32x8]: Sleef_tanhf8_u10avx2); - impl_unary!(f64x8[h => f64x4]: Sleef_tanhd4_u10avx2); - - impl_unary!(f32x4: Sleef_tanhf4_u10avx2128); - impl_unary!(f32x8: Sleef_tanhf8_u10avx2); - impl_unary!(f64x2: Sleef_tanhd2_u10avx2128); - impl_unary!(f64x4: Sleef_tanhd4_u10avx2); - } else if #[cfg(target_feature = "avx")] { - impl_unary!(f32x2[t => f32x4]: Sleef_tanhf4_u10sse4); - impl_unary!(f32x16[h => f32x8]: Sleef_tanhf8_u10avx); - impl_unary!(f64x8[h => f64x4]: Sleef_tanhd4_u10avx); - - impl_unary!(f32x4: Sleef_tanhf4_u10sse4); - impl_unary!(f32x8: Sleef_tanhf8_u10avx); - impl_unary!(f64x2: Sleef_tanhd2_u10sse4); - impl_unary!(f64x4: Sleef_tanhd4_u10avx); - } else if #[cfg(target_feature = "sse4.2")] { - impl_unary!(f32x2[t => f32x4]: Sleef_tanhf4_u10sse4); - impl_unary!(f32x16[q => f32x4]: Sleef_tanhf4_u10sse4); - impl_unary!(f64x8[q => f64x2]: Sleef_tanhd2_u10sse4); - - impl_unary!(f32x4: Sleef_tanhf4_u10sse4); - impl_unary!(f32x8[h => f32x4]: Sleef_tanhf4_u10sse4); - impl_unary!(f64x2: Sleef_tanhd2_u10sse4); - impl_unary!(f64x4[h => f64x2]: Sleef_tanhd2_u10sse4); - } else { - impl_unary!(f32x2[f32; 2]: tanh_f32); - impl_unary!(f32x16: tanh_v16f32); - impl_unary!(f64x8: tanh_v8f64); - - impl_unary!(f32x4: tanh_v4f32); - impl_unary!(f32x8: tanh_v8f32); - impl_unary!(f64x2: tanh_v2f64); - impl_unary!(f64x4: tanh_v4f64); - } - } - } else { - impl_unary!(f32x2[f32; 2]: tanh_f32); - impl_unary!(f32x4: tanh_v4f32); - impl_unary!(f32x8: tanh_v8f32); - impl_unary!(f32x16: tanh_v16f32); - - impl_unary!(f64x2: tanh_v2f64); - impl_unary!(f64x4: tanh_v4f64); - impl_unary!(f64x8: tanh_v8f64); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs b/third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs deleted file mode 100644 index 55cbc297aa..0000000000 --- a/third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs +++ /dev/null @@ -1,28 +0,0 @@ -//! Provides `isize` and `usize` - -use cfg_if::cfg_if; - -cfg_if! { - if #[cfg(target_pointer_width = "8")] { - pub(crate) type isize_ = i8; - pub(crate) type usize_ = u8; - } else if #[cfg(target_pointer_width = "16")] { - pub(crate) type isize_ = i16; - pub(crate) type usize_ = u16; - } else if #[cfg(target_pointer_width = "32")] { - pub(crate) type isize_ = i32; - pub(crate) type usize_ = u32; - - } else if #[cfg(target_pointer_width = "64")] { - pub(crate) type isize_ = i64; - pub(crate) type usize_ = u64; - } else if #[cfg(target_pointer_width = "64")] { - pub(crate) type isize_ = i64; - pub(crate) type usize_ = u64; - } else if #[cfg(target_pointer_width = "128")] { - pub(crate) type isize_ = i128; - pub(crate) type usize_ = u128; - } else { - compile_error!("unsupported target_pointer_width"); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions.rs b/third_party/rust/packed_simd/src/codegen/reductions.rs deleted file mode 100644 index 302ca6d88f..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions.rs +++ /dev/null @@ -1 +0,0 @@ -pub(crate) mod mask; diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask.rs deleted file mode 100644 index a78bcc5632..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask.rs +++ /dev/null @@ -1,69 +0,0 @@ -//! Code generation workaround for `all()` mask horizontal reduction. -//! -//! Works around [LLVM bug 36702]. -//! -//! [LLVM bug 36702]: https://bugs.llvm.org/show_bug.cgi?id=36702 -#![allow(unused_macros)] - -use crate::*; - -pub(crate) trait All: crate::marker::Sized { - unsafe fn all(self) -> bool; -} - -pub(crate) trait Any: crate::marker::Sized { - unsafe fn any(self) -> bool; -} - -#[macro_use] -mod fallback_impl; - -cfg_if! { - if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] { - #[macro_use] - mod x86; - } else if #[cfg(all(target_arch = "arm", target_feature = "v7", - target_feature = "neon", - any(feature = "core_arch", libcore_neon)))] { - #[macro_use] - mod arm; - } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon"))] { - #[macro_use] - mod aarch64; - } else { - #[macro_use] - mod fallback; - } -} - -impl_mask_reductions!(m8x2); -impl_mask_reductions!(m8x4); -impl_mask_reductions!(m8x8); -impl_mask_reductions!(m8x16); -impl_mask_reductions!(m8x32); -impl_mask_reductions!(m8x64); - -impl_mask_reductions!(m16x2); -impl_mask_reductions!(m16x4); -impl_mask_reductions!(m16x8); -impl_mask_reductions!(m16x16); -impl_mask_reductions!(m16x32); - -impl_mask_reductions!(m32x2); -impl_mask_reductions!(m32x4); -impl_mask_reductions!(m32x8); -impl_mask_reductions!(m32x16); - -// FIXME: 64-bit single element vector -// impl_mask_reductions!(m64x1); -impl_mask_reductions!(m64x2); -impl_mask_reductions!(m64x4); -impl_mask_reductions!(m64x8); - -impl_mask_reductions!(m128x1); -impl_mask_reductions!(m128x2); -impl_mask_reductions!(m128x4); - -impl_mask_reductions!(msizex2); -impl_mask_reductions!(msizex4); -impl_mask_reductions!(msizex8); diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs deleted file mode 100644 index b2db52c891..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs +++ /dev/null @@ -1,81 +0,0 @@ -//! Mask reductions implementation for `aarch64` targets - -/// 128-bit wide vectors -macro_rules! aarch64_128_neon_impl { - ($id:ident, $vmin:ident, $vmax:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "neon")] - unsafe fn all(self) -> bool { - use crate::arch::aarch64::$vmin; - $vmin(crate::mem::transmute(self)) != 0 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "neon")] - unsafe fn any(self) -> bool { - use crate::arch::aarch64::$vmax; - $vmax(crate::mem::transmute(self)) != 0 - } - } - }; -} - -/// 64-bit wide vectors -macro_rules! aarch64_64_neon_impl { - ($id:ident, $vec128:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "neon")] - unsafe fn all(self) -> bool { - // Duplicates the 64-bit vector into a 128-bit one and - // calls all on that. - union U { - halves: ($id, $id), - vec: $vec128, - } - U { halves: (self, self) }.vec.all() - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "neon")] - unsafe fn any(self) -> bool { - union U { - halves: ($id, $id), - vec: $vec128, - } - U { halves: (self, self) }.vec.any() - } - } - }; -} - -/// Mask reduction implementation for `aarch64` targets -macro_rules! impl_mask_reductions { - // 64-bit wide masks - (m8x8) => { - aarch64_64_neon_impl!(m8x8, m8x16); - }; - (m16x4) => { - aarch64_64_neon_impl!(m16x4, m16x8); - }; - (m32x2) => { - aarch64_64_neon_impl!(m32x2, m32x4); - }; - // 128-bit wide masks - (m8x16) => { - aarch64_128_neon_impl!(m8x16, vminvq_u8, vmaxvq_u8); - }; - (m16x8) => { - aarch64_128_neon_impl!(m16x8, vminvq_u16, vmaxvq_u16); - }; - (m32x4) => { - aarch64_128_neon_impl!(m32x4, vminvq_u32, vmaxvq_u32); - }; - // Fallback to LLVM's default code-generation: - ($id:ident) => { - fallback_impl!($id); - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs deleted file mode 100644 index 41c3cbc58a..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs +++ /dev/null @@ -1,56 +0,0 @@ -//! Mask reductions implementation for `arm` targets - -/// Implementation for ARM + v7 + NEON for 64-bit or 128-bit wide vectors with -/// more than two elements. -macro_rules! arm_128_v7_neon_impl { - ($id:ident, $half:ident, $vpmin:ident, $vpmax:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "v7,neon")] - unsafe fn all(self) -> bool { - use crate::arch::arm::$vpmin; - use crate::mem::transmute; - union U { - halves: ($half, $half), - vec: $id, - } - let halves = U { vec: self }.halves; - let h: $half = transmute($vpmin(transmute(halves.0), transmute(halves.1))); - h.all() - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "v7,neon")] - unsafe fn any(self) -> bool { - use crate::arch::arm::$vpmax; - use crate::mem::transmute; - union U { - halves: ($half, $half), - vec: $id, - } - let halves = U { vec: self }.halves; - let h: $half = transmute($vpmax(transmute(halves.0), transmute(halves.1))); - h.any() - } - } - }; -} - -/// Mask reduction implementation for `arm` targets -macro_rules! impl_mask_reductions { - // 128-bit wide masks - (m8x16) => { - arm_128_v7_neon_impl!(m8x16, m8x8, vpmin_u8, vpmax_u8); - }; - (m16x8) => { - arm_128_v7_neon_impl!(m16x8, m16x4, vpmin_u16, vpmax_u16); - }; - (m32x4) => { - arm_128_v7_neon_impl!(m32x4, m32x2, vpmin_u32, vpmax_u32); - }; - // Fallback to LLVM's default code-generation: - ($id:ident) => { - fallback_impl!($id); - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs deleted file mode 100644 index 4c377a6878..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs +++ /dev/null @@ -1,8 +0,0 @@ -//! Default mask reduction implementations. - -/// Default mask reduction implementation -macro_rules! impl_mask_reductions { - ($id:ident) => { - fallback_impl!($id); - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs deleted file mode 100644 index 0d246e2fda..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs +++ /dev/null @@ -1,237 +0,0 @@ -//! Default implementation of a mask reduction for any target. - -macro_rules! fallback_to_other_impl { - ($id:ident, $other:ident) => { - impl All for $id { - #[inline] - unsafe fn all(self) -> bool { - let m: $other = crate::mem::transmute(self); - m.all() - } - } - impl Any for $id { - #[inline] - unsafe fn any(self) -> bool { - let m: $other = crate::mem::transmute(self); - m.any() - } - } - }; -} - -/// Fallback implementation. -macro_rules! fallback_impl { - // 16-bit wide masks: - (m8x2) => { - impl All for m8x2 { - #[inline] - unsafe fn all(self) -> bool { - let i: u16 = crate::mem::transmute(self); - i == u16::max_value() - } - } - impl Any for m8x2 { - #[inline] - unsafe fn any(self) -> bool { - let i: u16 = crate::mem::transmute(self); - i != 0 - } - } - }; - // 32-bit wide masks - (m8x4) => { - impl All for m8x4 { - #[inline] - unsafe fn all(self) -> bool { - let i: u32 = crate::mem::transmute(self); - i == u32::max_value() - } - } - impl Any for m8x4 { - #[inline] - unsafe fn any(self) -> bool { - let i: u32 = crate::mem::transmute(self); - i != 0 - } - } - }; - (m16x2) => { - fallback_to_other_impl!(m16x2, m8x4); - }; - // 64-bit wide masks: - (m8x8) => { - impl All for m8x8 { - #[inline] - unsafe fn all(self) -> bool { - let i: u64 = crate::mem::transmute(self); - i == u64::max_value() - } - } - impl Any for m8x8 { - #[inline] - unsafe fn any(self) -> bool { - let i: u64 = crate::mem::transmute(self); - i != 0 - } - } - }; - (m16x4) => { - fallback_to_other_impl!(m16x4, m8x8); - }; - (m32x2) => { - fallback_to_other_impl!(m32x2, m16x4); - }; - // FIXME: 64x1 maxk - // 128-bit wide masks: - (m8x16) => { - impl All for m8x16 { - #[inline] - unsafe fn all(self) -> bool { - let i: u128 = crate::mem::transmute(self); - i == u128::max_value() - } - } - impl Any for m8x16 { - #[inline] - unsafe fn any(self) -> bool { - let i: u128 = crate::mem::transmute(self); - i != 0 - } - } - }; - (m16x8) => { - fallback_to_other_impl!(m16x8, m8x16); - }; - (m32x4) => { - fallback_to_other_impl!(m32x4, m16x8); - }; - (m64x2) => { - fallback_to_other_impl!(m64x2, m32x4); - }; - (m128x1) => { - fallback_to_other_impl!(m128x1, m64x2); - }; - // 256-bit wide masks - (m8x32) => { - impl All for m8x32 { - #[inline] - unsafe fn all(self) -> bool { - let i: [u128; 2] = crate::mem::transmute(self); - let o: [u128; 2] = [u128::max_value(); 2]; - i == o - } - } - impl Any for m8x32 { - #[inline] - unsafe fn any(self) -> bool { - let i: [u128; 2] = crate::mem::transmute(self); - let o: [u128; 2] = [0; 2]; - i != o - } - } - }; - (m16x16) => { - fallback_to_other_impl!(m16x16, m8x32); - }; - (m32x8) => { - fallback_to_other_impl!(m32x8, m16x16); - }; - (m64x4) => { - fallback_to_other_impl!(m64x4, m32x8); - }; - (m128x2) => { - fallback_to_other_impl!(m128x2, m64x4); - }; - // 512-bit wide masks - (m8x64) => { - impl All for m8x64 { - #[inline] - unsafe fn all(self) -> bool { - let i: [u128; 4] = crate::mem::transmute(self); - let o: [u128; 4] = [u128::max_value(); 4]; - i == o - } - } - impl Any for m8x64 { - #[inline] - unsafe fn any(self) -> bool { - let i: [u128; 4] = crate::mem::transmute(self); - let o: [u128; 4] = [0; 4]; - i != o - } - } - }; - (m16x32) => { - fallback_to_other_impl!(m16x32, m8x64); - }; - (m32x16) => { - fallback_to_other_impl!(m32x16, m16x32); - }; - (m64x8) => { - fallback_to_other_impl!(m64x8, m32x16); - }; - (m128x4) => { - fallback_to_other_impl!(m128x4, m64x8); - }; - // Masks with pointer-sized elements64 - (msizex2) => { - cfg_if! { - if #[cfg(target_pointer_width = "64")] { - fallback_to_other_impl!(msizex2, m64x2); - } else if #[cfg(target_pointer_width = "32")] { - fallback_to_other_impl!(msizex2, m32x2); - } else { - compile_error!("unsupported target_pointer_width"); - } - } - }; - (msizex4) => { - cfg_if! { - if #[cfg(target_pointer_width = "64")] { - fallback_to_other_impl!(msizex4, m64x4); - } else if #[cfg(target_pointer_width = "32")] { - fallback_to_other_impl!(msizex4, m32x4); - } else { - compile_error!("unsupported target_pointer_width"); - } - } - }; - (msizex8) => { - cfg_if! { - if #[cfg(target_pointer_width = "64")] { - fallback_to_other_impl!(msizex8, m64x8); - } else if #[cfg(target_pointer_width = "32")] { - fallback_to_other_impl!(msizex8, m32x8); - } else { - compile_error!("unsupported target_pointer_width"); - } - } - }; -} - -macro_rules! recurse_half { - ($vid:ident, $vid_h:ident) => { - impl All for $vid { - #[inline] - unsafe fn all(self) -> bool { - union U { - halves: ($vid_h, $vid_h), - vec: $vid, - } - let halves = U { vec: self }.halves; - halves.0.all() && halves.1.all() - } - } - impl Any for $vid { - #[inline] - unsafe fn any(self) -> bool { - union U { - halves: ($vid_h, $vid_h), - vec: $vid, - } - let halves = U { vec: self }.halves; - halves.0.any() || halves.1.any() - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs deleted file mode 100644 index 4bf5098065..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs +++ /dev/null @@ -1,216 +0,0 @@ -//! Mask reductions implementation for `x86` and `x86_64` targets - -#[cfg(target_feature = "sse")] -#[macro_use] -mod sse; - -#[cfg(target_feature = "sse2")] -#[macro_use] -mod sse2; - -#[cfg(target_feature = "avx")] -#[macro_use] -mod avx; - -#[cfg(target_feature = "avx2")] -#[macro_use] -mod avx2; - -/// x86 64-bit m8x8 implementation -macro_rules! x86_m8x8_impl { - ($id:ident) => { - fallback_impl!($id); - }; -} - -/// x86 128-bit m8x16 implementation -macro_rules! x86_m8x16_impl { - ($id:ident) => { - cfg_if! { - if #[cfg(target_feature = "sse2")] { - x86_m8x16_sse2_impl!($id); - } else { - fallback_impl!($id); - } - } - }; -} - -/// x86 128-bit m32x4 implementation -macro_rules! x86_m32x4_impl { - ($id:ident) => { - cfg_if! { - if #[cfg(target_feature = "sse")] { - x86_m32x4_sse_impl!($id); - } else { - fallback_impl!($id); - } - } - }; -} - -/// x86 128-bit m64x2 implementation -macro_rules! x86_m64x2_impl { - ($id:ident) => { - cfg_if! { - if #[cfg(target_feature = "sse2")] { - x86_m64x2_sse2_impl!($id); - } else if #[cfg(target_feature = "sse")] { - x86_m32x4_sse_impl!($id); - } else { - fallback_impl!($id); - } - } - }; -} - -/// x86 256-bit m8x32 implementation -macro_rules! x86_m8x32_impl { - ($id:ident, $half_id:ident) => { - cfg_if! { - if #[cfg(target_feature = "avx2")] { - x86_m8x32_avx2_impl!($id); - } else if #[cfg(target_feature = "avx")] { - x86_m8x32_avx_impl!($id); - } else if #[cfg(target_feature = "sse2")] { - recurse_half!($id, $half_id); - } else { - fallback_impl!($id); - } - } - }; -} - -/// x86 256-bit m32x8 implementation -macro_rules! x86_m32x8_impl { - ($id:ident, $half_id:ident) => { - cfg_if! { - if #[cfg(target_feature = "avx")] { - x86_m32x8_avx_impl!($id); - } else if #[cfg(target_feature = "sse")] { - recurse_half!($id, $half_id); - } else { - fallback_impl!($id); - } - } - }; -} - -/// x86 256-bit m64x4 implementation -macro_rules! x86_m64x4_impl { - ($id:ident, $half_id:ident) => { - cfg_if! { - if #[cfg(target_feature = "avx")] { - x86_m64x4_avx_impl!($id); - } else if #[cfg(target_feature = "sse")] { - recurse_half!($id, $half_id); - } else { - fallback_impl!($id); - } - } - }; -} - -/// Fallback implementation. -macro_rules! x86_intr_impl { - ($id:ident) => { - impl All for $id { - #[inline] - unsafe fn all(self) -> bool { - use crate::llvm::simd_reduce_all; - simd_reduce_all(self.0) - } - } - impl Any for $id { - #[inline] - unsafe fn any(self) -> bool { - use crate::llvm::simd_reduce_any; - simd_reduce_any(self.0) - } - } - }; -} - -/// Mask reduction implementation for `x86` and `x86_64` targets -macro_rules! impl_mask_reductions { - // 64-bit wide masks - (m8x8) => { - x86_m8x8_impl!(m8x8); - }; - (m16x4) => { - x86_m8x8_impl!(m16x4); - }; - (m32x2) => { - x86_m8x8_impl!(m32x2); - }; - // 128-bit wide masks - (m8x16) => { - x86_m8x16_impl!(m8x16); - }; - (m16x8) => { - x86_m8x16_impl!(m16x8); - }; - (m32x4) => { - x86_m32x4_impl!(m32x4); - }; - (m64x2) => { - x86_m64x2_impl!(m64x2); - }; - (m128x1) => { - x86_intr_impl!(m128x1); - }; - // 256-bit wide masks: - (m8x32) => { - x86_m8x32_impl!(m8x32, m8x16); - }; - (m16x16) => { - x86_m8x32_impl!(m16x16, m16x8); - }; - (m32x8) => { - x86_m32x8_impl!(m32x8, m32x4); - }; - (m64x4) => { - x86_m64x4_impl!(m64x4, m64x2); - }; - (m128x2) => { - x86_intr_impl!(m128x2); - }; - (msizex2) => { - cfg_if! { - if #[cfg(target_pointer_width = "64")] { - fallback_to_other_impl!(msizex2, m64x2); - } else if #[cfg(target_pointer_width = "32")] { - fallback_to_other_impl!(msizex2, m32x2); - } else { - compile_error!("unsupported target_pointer_width"); - } - } - }; - (msizex4) => { - cfg_if! { - if #[cfg(target_pointer_width = "64")] { - fallback_to_other_impl!(msizex4, m64x4); - } else if #[cfg(target_pointer_width = "32")] { - fallback_to_other_impl!(msizex4, m32x4); - } else { - compile_error!("unsupported target_pointer_width"); - } - } - }; - (msizex8) => { - cfg_if! { - if #[cfg(target_pointer_width = "64")] { - fallback_to_other_impl!(msizex8, m64x8); - } else if #[cfg(target_pointer_width = "32")] { - fallback_to_other_impl!(msizex8, m32x8); - } else { - compile_error!("unsupported target_pointer_width"); - } - } - }; - - // Fallback to LLVM's default code-generation: - ($id:ident) => { - fallback_impl!($id); - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs deleted file mode 100644 index 61f352d228..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs +++ /dev/null @@ -1,95 +0,0 @@ -//! Mask reductions implementation for `x86` and `x86_64` targets with `AVX` - -/// `x86`/`x86_64` 256-bit `AVX` implementation -/// FIXME: it might be faster here to do two `_mm_movmask_epi8` -#[cfg(target_feature = "avx")] -macro_rules! x86_m8x32_avx_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "avx")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_testc_si256; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_testc_si256; - _mm256_testc_si256(crate::mem::transmute(self), crate::mem::transmute($id::splat(true))) != 0 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "avx")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_testz_si256; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_testz_si256; - _mm256_testz_si256(crate::mem::transmute(self), crate::mem::transmute(self)) == 0 - } - } - }; -} - -/// `x86`/`x86_64` 256-bit m32x8 `AVX` implementation -macro_rules! x86_m32x8_avx_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_movemask_ps; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_movemask_ps; - // _mm256_movemask_ps(a) creates a 8bit mask containing the - // most significant bit of each lane of `a`. If all bits are - // set, then all 8 lanes of the mask are true. - _mm256_movemask_ps(crate::mem::transmute(self)) == 0b_1111_1111_i32 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_movemask_ps; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_movemask_ps; - - _mm256_movemask_ps(crate::mem::transmute(self)) != 0 - } - } - }; -} - -/// `x86`/`x86_64` 256-bit m64x4 `AVX` implementation -macro_rules! x86_m64x4_avx_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_movemask_pd; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_movemask_pd; - // _mm256_movemask_pd(a) creates a 4bit mask containing the - // most significant bit of each lane of `a`. If all bits are - // set, then all 4 lanes of the mask are true. - _mm256_movemask_pd(crate::mem::transmute(self)) == 0b_1111_i32 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_movemask_pd; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_movemask_pd; - - _mm256_movemask_pd(crate::mem::transmute(self)) != 0 - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs deleted file mode 100644 index d37d023420..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs +++ /dev/null @@ -1,35 +0,0 @@ -//! Mask reductions implementation for `x86` and `x86_64` targets with `AVX2`. -#![allow(unused)] - -/// x86/x86_64 256-bit m8x32 AVX2 implementation -macro_rules! x86_m8x32_avx2_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "sse2")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_movemask_epi8; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_movemask_epi8; - // _mm256_movemask_epi8(a) creates a 32bit mask containing the - // most significant bit of each byte of `a`. If all - // bits are set, then all 32 lanes of the mask are - // true. - _mm256_movemask_epi8(crate::mem::transmute(self)) == -1_i32 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "sse2")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm256_movemask_epi8; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm256_movemask_epi8; - - _mm256_movemask_epi8(crate::mem::transmute(self)) != 0 - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs deleted file mode 100644 index e0c9aee92b..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs +++ /dev/null @@ -1,35 +0,0 @@ -//! Mask reductions implementation for `x86` and `x86_64` targets with `SSE`. -#![allow(unused)] - -/// `x86`/`x86_64` 128-bit `m32x4` `SSE` implementation -macro_rules! x86_m32x4_sse_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_movemask_ps; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_movemask_ps; - // _mm_movemask_ps(a) creates a 4bit mask containing the - // most significant bit of each lane of `a`. If all - // bits are set, then all 4 lanes of the mask are - // true. - _mm_movemask_ps(crate::mem::transmute(self)) == 0b_1111_i32 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_movemask_ps; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_movemask_ps; - - _mm_movemask_ps(crate::mem::transmute(self)) != 0 - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs deleted file mode 100644 index bbb52fa47e..0000000000 --- a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs +++ /dev/null @@ -1,68 +0,0 @@ -//! Mask reductions implementation for `x86` and `x86_64` targets with `SSE2`. -#![allow(unused)] - -/// `x86`/`x86_64` 128-bit m64x2 `SSE2` implementation -macro_rules! x86_m64x2_sse2_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_movemask_pd; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_movemask_pd; - // _mm_movemask_pd(a) creates a 2bit mask containing the - // most significant bit of each lane of `a`. If all - // bits are set, then all 2 lanes of the mask are - // true. - _mm_movemask_pd(crate::mem::transmute(self)) == 0b_11_i32 - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "sse")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_movemask_pd; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_movemask_pd; - - _mm_movemask_pd(crate::mem::transmute(self)) != 0 - } - } - }; -} - -/// `x86`/`x86_64` 128-bit m8x16 `SSE2` implementation -macro_rules! x86_m8x16_sse2_impl { - ($id:ident) => { - impl All for $id { - #[inline] - #[target_feature(enable = "sse2")] - unsafe fn all(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_movemask_epi8; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_movemask_epi8; - // _mm_movemask_epi8(a) creates a 16bit mask containing the - // most significant bit of each byte of `a`. If all - // bits are set, then all 16 lanes of the mask are - // true. - _mm_movemask_epi8(crate::mem::transmute(self)) == i32::from(u16::max_value()) - } - } - impl Any for $id { - #[inline] - #[target_feature(enable = "sse2")] - unsafe fn any(self) -> bool { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_movemask_epi8; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_movemask_epi8; - - _mm_movemask_epi8(crate::mem::transmute(self)) != 0 - } - } - }; -} diff --git a/third_party/rust/packed_simd/src/codegen/shuffle.rs b/third_party/rust/packed_simd/src/codegen/shuffle.rs deleted file mode 100644 index d3acd48f5b..0000000000 --- a/third_party/rust/packed_simd/src/codegen/shuffle.rs +++ /dev/null @@ -1,150 +0,0 @@ -//! Implementations of the `ShuffleResult` trait for the different numbers of -//! lanes and vector element types. - -use crate::masks::*; -use crate::sealed::{Seal, Shuffle}; - -macro_rules! impl_shuffle { - ($array:ty, $base:ty, $out:ty) => { - impl Seal<$array> for $base {} - impl Shuffle<$array> for $base { - type Output = $out; - } - }; -} - -impl_shuffle! { [u32; 2], i8, crate::codegen::i8x2 } -impl_shuffle! { [u32; 4], i8, crate::codegen::i8x4 } -impl_shuffle! { [u32; 8], i8, crate::codegen::i8x8 } -impl_shuffle! { [u32; 16], i8, crate::codegen::i8x16 } -impl_shuffle! { [u32; 32], i8, crate::codegen::i8x32 } -impl_shuffle! { [u32; 64], i8, crate::codegen::i8x64 } - -impl_shuffle! { [u32; 2], u8, crate::codegen::u8x2 } -impl_shuffle! { [u32; 4], u8, crate::codegen::u8x4 } -impl_shuffle! { [u32; 8], u8, crate::codegen::u8x8 } -impl_shuffle! { [u32; 16], u8, crate::codegen::u8x16 } -impl_shuffle! { [u32; 32], u8, crate::codegen::u8x32 } -impl_shuffle! { [u32; 64], u8, crate::codegen::u8x64 } - -impl_shuffle! { [u32; 2], m8, crate::codegen::m8x2 } -impl_shuffle! { [u32; 4], m8, crate::codegen::m8x4 } -impl_shuffle! { [u32; 8], m8, crate::codegen::m8x8 } -impl_shuffle! { [u32; 16], m8, crate::codegen::m8x16 } -impl_shuffle! { [u32; 32], m8, crate::codegen::m8x32 } -impl_shuffle! { [u32; 64], m8, crate::codegen::m8x64 } - -impl_shuffle! { [u32; 2], i16, crate::codegen::i16x2 } -impl_shuffle! { [u32; 4], i16, crate::codegen::i16x4 } -impl_shuffle! { [u32; 8], i16, crate::codegen::i16x8 } -impl_shuffle! { [u32; 16], i16, crate::codegen::i16x16 } -impl_shuffle! { [u32; 32], i16, crate::codegen::i16x32 } - -impl_shuffle! { [u32; 2], u16, crate::codegen::u16x2 } -impl_shuffle! { [u32; 4], u16, crate::codegen::u16x4 } -impl_shuffle! { [u32; 8], u16, crate::codegen::u16x8 } -impl_shuffle! { [u32; 16], u16, crate::codegen::u16x16 } -impl_shuffle! { [u32; 32], u16, crate::codegen::u16x32 } - -impl_shuffle! { [u32; 2], m16, crate::codegen::m16x2 } -impl_shuffle! { [u32; 4], m16, crate::codegen::m16x4 } -impl_shuffle! { [u32; 8], m16, crate::codegen::m16x8 } -impl_shuffle! { [u32; 16], m16, crate::codegen::m16x16 } - -impl_shuffle! { [u32; 2], i32, crate::codegen::i32x2 } -impl_shuffle! { [u32; 4], i32, crate::codegen::i32x4 } -impl_shuffle! { [u32; 8], i32, crate::codegen::i32x8 } -impl_shuffle! { [u32; 16], i32, crate::codegen::i32x16 } - -impl_shuffle! { [u32; 2], u32, crate::codegen::u32x2 } -impl_shuffle! { [u32; 4], u32, crate::codegen::u32x4 } -impl_shuffle! { [u32; 8], u32, crate::codegen::u32x8 } -impl_shuffle! { [u32; 16], u32, crate::codegen::u32x16 } - -impl_shuffle! { [u32; 2], f32, crate::codegen::f32x2 } -impl_shuffle! { [u32; 4], f32, crate::codegen::f32x4 } -impl_shuffle! { [u32; 8], f32, crate::codegen::f32x8 } -impl_shuffle! { [u32; 16], f32, crate::codegen::f32x16 } - -impl_shuffle! { [u32; 2], m32, crate::codegen::m32x2 } -impl_shuffle! { [u32; 4], m32, crate::codegen::m32x4 } -impl_shuffle! { [u32; 8], m32, crate::codegen::m32x8 } -impl_shuffle! { [u32; 16], m32, crate::codegen::m32x16 } - -/* FIXME: 64-bit single element vector -impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } -*/ -impl_shuffle! { [u32; 2], i64, crate::codegen::i64x2 } -impl_shuffle! { [u32; 4], i64, crate::codegen::i64x4 } -impl_shuffle! { [u32; 8], i64, crate::codegen::i64x8 } - -/* FIXME: 64-bit single element vector -impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } -*/ -impl_shuffle! { [u32; 2], u64, crate::codegen::u64x2 } -impl_shuffle! { [u32; 4], u64, crate::codegen::u64x4 } -impl_shuffle! { [u32; 8], u64, crate::codegen::u64x8 } - -/* FIXME: 64-bit single element vector -impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } -*/ -impl_shuffle! { [u32; 2], f64, crate::codegen::f64x2 } -impl_shuffle! { [u32; 4], f64, crate::codegen::f64x4 } -impl_shuffle! { [u32; 8], f64, crate::codegen::f64x8 } - -/* FIXME: 64-bit single element vector -impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } -*/ -impl_shuffle! { [u32; 2], m64, crate::codegen::m64x2 } -impl_shuffle! { [u32; 4], m64, crate::codegen::m64x4 } -impl_shuffle! { [u32; 8], m64, crate::codegen::m64x8 } - -impl_shuffle! { [u32; 2], isize, crate::codegen::isizex2 } -impl_shuffle! { [u32; 4], isize, crate::codegen::isizex4 } -impl_shuffle! { [u32; 8], isize, crate::codegen::isizex8 } - -impl_shuffle! { [u32; 2], usize, crate::codegen::usizex2 } -impl_shuffle! { [u32; 4], usize, crate::codegen::usizex4 } -impl_shuffle! { [u32; 8], usize, crate::codegen::usizex8 } - -impl_shuffle! { [u32; 2], msize, crate::codegen::msizex2 } -impl_shuffle! { [u32; 4], msize, crate::codegen::msizex4 } -impl_shuffle! { [u32; 8], msize, crate::codegen::msizex8 } - -impl<T> Seal<[u32; 2]> for *const T {} -impl<T> Shuffle<[u32; 2]> for *const T { - type Output = crate::codegen::cptrx2<T>; -} -impl<T> Seal<[u32; 4]> for *const T {} -impl<T> Shuffle<[u32; 4]> for *const T { - type Output = crate::codegen::cptrx4<T>; -} -impl<T> Seal<[u32; 8]> for *const T {} -impl<T> Shuffle<[u32; 8]> for *const T { - type Output = crate::codegen::cptrx8<T>; -} - -impl<T> Seal<[u32; 2]> for *mut T {} -impl<T> Shuffle<[u32; 2]> for *mut T { - type Output = crate::codegen::mptrx2<T>; -} -impl<T> Seal<[u32; 4]> for *mut T {} -impl<T> Shuffle<[u32; 4]> for *mut T { - type Output = crate::codegen::mptrx4<T>; -} -impl<T> Seal<[u32; 8]> for *mut T {} -impl<T> Shuffle<[u32; 8]> for *mut T { - type Output = crate::codegen::mptrx8<T>; -} - -impl_shuffle! { [u32; 1], i128, crate::codegen::i128x1 } -impl_shuffle! { [u32; 2], i128, crate::codegen::i128x2 } -impl_shuffle! { [u32; 4], i128, crate::codegen::i128x4 } - -impl_shuffle! { [u32; 1], u128, crate::codegen::u128x1 } -impl_shuffle! { [u32; 2], u128, crate::codegen::u128x2 } -impl_shuffle! { [u32; 4], u128, crate::codegen::u128x4 } - -impl_shuffle! { [u32; 1], m128, crate::codegen::m128x1 } -impl_shuffle! { [u32; 2], m128, crate::codegen::m128x2 } -impl_shuffle! { [u32; 4], m128, crate::codegen::m128x4 } diff --git a/third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs b/third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs deleted file mode 100644 index 19d457a45b..0000000000 --- a/third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs +++ /dev/null @@ -1,408 +0,0 @@ -//! Shuffle vector lanes with run-time indices. - -use crate::*; - -pub trait Shuffle1Dyn { - type Indices; - fn shuffle1_dyn(self, _: Self::Indices) -> Self; -} - -// Fallback implementation -macro_rules! impl_fallback { - ($id:ident) => { - impl Shuffle1Dyn for $id { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - let mut result = Self::splat(0); - for i in 0..$id::lanes() { - result = result.replace(i, self.extract(indices.extract(i) as usize)); - } - result - } - } - }; -} - -macro_rules! impl_shuffle1_dyn { - (u8x8) => { - cfg_if! { - if #[cfg(all( - any( - all(target_arch = "aarch64", target_feature = "neon"), - all(target_arch = "doesnotexist", target_feature = "v7", - target_feature = "neon") - ), - any(feature = "core_arch", libcore_neon) - ) - )] { - impl Shuffle1Dyn for u8x8 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - #[cfg(target_arch = "aarch64")] - use crate::arch::aarch64::vtbl1_u8; - #[cfg(target_arch = "doesnotexist")] - use crate::arch::arm::vtbl1_u8; - - // This is safe because the binary is compiled with - // neon enabled at compile-time and can therefore only - // run on CPUs that have it enabled. - unsafe { - Simd(mem::transmute( - vtbl1_u8(mem::transmute(self.0), - crate::mem::transmute(indices.0)) - )) - } - } - } - } else { - impl_fallback!(u8x8); - } - } - }; - (u8x16) => { - cfg_if! { - if #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), - target_feature = "ssse3"))] { - impl Shuffle1Dyn for u8x16 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - #[cfg(target_arch = "x86")] - use crate::arch::x86::_mm_shuffle_epi8; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::_mm_shuffle_epi8; - // This is safe because the binary is compiled with - // ssse3 enabled at compile-time and can therefore only - // run on CPUs that have it enabled. - unsafe { - Simd(mem::transmute( - _mm_shuffle_epi8(mem::transmute(self.0), - crate::mem::transmute(indices)) - )) - } - } - } - } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon", - any(feature = "core_arch", libcore_neon)))] { - impl Shuffle1Dyn for u8x16 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - use crate::arch::aarch64::vqtbl1q_u8; - - // This is safe because the binary is compiled with - // neon enabled at compile-time and can therefore only - // run on CPUs that have it enabled. - unsafe { - Simd(mem::transmute( - vqtbl1q_u8(mem::transmute(self.0), - crate::mem::transmute(indices.0)) - )) - } - } - } - } else if #[cfg(all(target_arch = "doesnotexist", target_feature = "v7", - target_feature = "neon", - any(feature = "core_arch", libcore_neon)))] { - impl Shuffle1Dyn for u8x16 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - use crate::arch::arm::vtbl2_u8; - - // This is safe because the binary is compiled with - // neon enabled at compile-time and can therefore only - // run on CPUs that have it enabled. - unsafe { - union U { - j: u8x16, - s: (u8x8, u8x8), - } - - let (i0, i1) = U { j: y }.s; - - let r0 = vtbl2_u8( - mem::transmute(x), - crate::mem::transmute(i0) - ); - let r1 = vtbl2_u8( - mem::transmute(x), - crate::mem::transmute(i1) - ); - - let r = U { s: (r0, r1) }.j; - - Simd(mem::transmute(r)) - } - } - } - } else { - impl_fallback!(u8x16); - } - } - }; - (u16x8) => { - impl Shuffle1Dyn for u16x8 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - let indices: u8x8 = (indices * 2).cast(); - let indices: u8x16 = shuffle!(indices, [0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]); - let v = u8x16::new(0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1); - let indices = indices + v; - unsafe { - let s: u8x16 = crate::mem::transmute(self); - crate::mem::transmute(s.shuffle1_dyn(indices)) - } - } - } - }; - (u32x4) => { - cfg_if! { - if #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), - target_feature = "avx"))] { - impl Shuffle1Dyn for u32x4 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - #[cfg(target_arch = "x86")] - use crate::arch::x86::{_mm_permutevar_ps}; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::{_mm_permutevar_ps}; - - unsafe { - crate::mem::transmute( - _mm_permutevar_ps( - crate::mem::transmute(self.0), - crate::mem::transmute(indices.0) - ) - ) - } - } - } - } else { - impl Shuffle1Dyn for u32x4 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - let indices: u8x4 = (indices * 4).cast(); - let indices: u8x16 = shuffle!( - indices, - [0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3] - ); - let v = u8x16::new( - 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3 - ); - let indices = indices + v; - unsafe { - let s: u8x16 =crate::mem::transmute(self); - crate::mem::transmute(s.shuffle1_dyn(indices)) - } - } - } - } - } - }; - (u64x2) => { - cfg_if! { - if #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), - target_feature = "avx"))] { - impl Shuffle1Dyn for u64x2 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - #[cfg(target_arch = "x86")] - use crate::arch::x86::{_mm_permutevar_pd}; - #[cfg(target_arch = "x86_64")] - use crate::arch::x86_64::{_mm_permutevar_pd}; - // _mm_permutevar_pd uses the _second_ bit of each - // element to perform the selection, that is: 0b00 => 0, - // 0b10 => 1: - let indices = indices << 1; - unsafe { - crate::mem::transmute( - _mm_permutevar_pd( - crate::mem::transmute(self), - crate::mem::transmute(indices) - ) - ) - } - } - } - } else { - impl Shuffle1Dyn for u64x2 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - let indices: u8x2 = (indices * 8).cast(); - let indices: u8x16 = shuffle!( - indices, - [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1] - ); - let v = u8x16::new( - 0, 1, 2, 3, 4, 5, 6, 7, 0, 1, 2, 3, 4, 5, 6, 7 - ); - let indices = indices + v; - unsafe { - let s: u8x16 =crate::mem::transmute(self); - crate::mem::transmute(s.shuffle1_dyn(indices)) - } - } - } - } - } - }; - (u128x1) => { - impl Shuffle1Dyn for u128x1 { - type Indices = Self; - #[inline] - fn shuffle1_dyn(self, _indices: Self::Indices) -> Self { - self - } - } - }; - ($id:ident) => { - impl_fallback!($id); - }; -} - -impl_shuffle1_dyn!(u8x2); -impl_shuffle1_dyn!(u8x4); -impl_shuffle1_dyn!(u8x8); -impl_shuffle1_dyn!(u8x16); -impl_shuffle1_dyn!(u8x32); -impl_shuffle1_dyn!(u8x64); - -impl_shuffle1_dyn!(u16x2); -impl_shuffle1_dyn!(u16x4); -impl_shuffle1_dyn!(u16x8); -impl_shuffle1_dyn!(u16x16); -impl_shuffle1_dyn!(u16x32); - -impl_shuffle1_dyn!(u32x2); -impl_shuffle1_dyn!(u32x4); -impl_shuffle1_dyn!(u32x8); -impl_shuffle1_dyn!(u32x16); - -impl_shuffle1_dyn!(u64x2); -impl_shuffle1_dyn!(u64x4); -impl_shuffle1_dyn!(u64x8); - -impl_shuffle1_dyn!(usizex2); -impl_shuffle1_dyn!(usizex4); -impl_shuffle1_dyn!(usizex8); - -impl_shuffle1_dyn!(u128x1); -impl_shuffle1_dyn!(u128x2); -impl_shuffle1_dyn!(u128x4); - -// Implementation for non-unsigned vector types -macro_rules! impl_shuffle1_dyn_non_u { - ($id:ident, $uid:ident) => { - impl Shuffle1Dyn for $id { - type Indices = $uid; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - unsafe { - let u: $uid = crate::mem::transmute(self); - crate::mem::transmute(u.shuffle1_dyn(indices)) - } - } - } - }; -} - -impl_shuffle1_dyn_non_u!(i8x2, u8x2); -impl_shuffle1_dyn_non_u!(i8x4, u8x4); -impl_shuffle1_dyn_non_u!(i8x8, u8x8); -impl_shuffle1_dyn_non_u!(i8x16, u8x16); -impl_shuffle1_dyn_non_u!(i8x32, u8x32); -impl_shuffle1_dyn_non_u!(i8x64, u8x64); - -impl_shuffle1_dyn_non_u!(i16x2, u16x2); -impl_shuffle1_dyn_non_u!(i16x4, u16x4); -impl_shuffle1_dyn_non_u!(i16x8, u16x8); -impl_shuffle1_dyn_non_u!(i16x16, u16x16); -impl_shuffle1_dyn_non_u!(i16x32, u16x32); - -impl_shuffle1_dyn_non_u!(i32x2, u32x2); -impl_shuffle1_dyn_non_u!(i32x4, u32x4); -impl_shuffle1_dyn_non_u!(i32x8, u32x8); -impl_shuffle1_dyn_non_u!(i32x16, u32x16); - -impl_shuffle1_dyn_non_u!(i64x2, u64x2); -impl_shuffle1_dyn_non_u!(i64x4, u64x4); -impl_shuffle1_dyn_non_u!(i64x8, u64x8); - -impl_shuffle1_dyn_non_u!(isizex2, usizex2); -impl_shuffle1_dyn_non_u!(isizex4, usizex4); -impl_shuffle1_dyn_non_u!(isizex8, usizex8); - -impl_shuffle1_dyn_non_u!(i128x1, u128x1); -impl_shuffle1_dyn_non_u!(i128x2, u128x2); -impl_shuffle1_dyn_non_u!(i128x4, u128x4); - -impl_shuffle1_dyn_non_u!(m8x2, u8x2); -impl_shuffle1_dyn_non_u!(m8x4, u8x4); -impl_shuffle1_dyn_non_u!(m8x8, u8x8); -impl_shuffle1_dyn_non_u!(m8x16, u8x16); -impl_shuffle1_dyn_non_u!(m8x32, u8x32); -impl_shuffle1_dyn_non_u!(m8x64, u8x64); - -impl_shuffle1_dyn_non_u!(m16x2, u16x2); -impl_shuffle1_dyn_non_u!(m16x4, u16x4); -impl_shuffle1_dyn_non_u!(m16x8, u16x8); -impl_shuffle1_dyn_non_u!(m16x16, u16x16); -impl_shuffle1_dyn_non_u!(m16x32, u16x32); - -impl_shuffle1_dyn_non_u!(m32x2, u32x2); -impl_shuffle1_dyn_non_u!(m32x4, u32x4); -impl_shuffle1_dyn_non_u!(m32x8, u32x8); -impl_shuffle1_dyn_non_u!(m32x16, u32x16); - -impl_shuffle1_dyn_non_u!(m64x2, u64x2); -impl_shuffle1_dyn_non_u!(m64x4, u64x4); -impl_shuffle1_dyn_non_u!(m64x8, u64x8); - -impl_shuffle1_dyn_non_u!(msizex2, usizex2); -impl_shuffle1_dyn_non_u!(msizex4, usizex4); -impl_shuffle1_dyn_non_u!(msizex8, usizex8); - -impl_shuffle1_dyn_non_u!(m128x1, u128x1); -impl_shuffle1_dyn_non_u!(m128x2, u128x2); -impl_shuffle1_dyn_non_u!(m128x4, u128x4); - -impl_shuffle1_dyn_non_u!(f32x2, u32x2); -impl_shuffle1_dyn_non_u!(f32x4, u32x4); -impl_shuffle1_dyn_non_u!(f32x8, u32x8); -impl_shuffle1_dyn_non_u!(f32x16, u32x16); - -impl_shuffle1_dyn_non_u!(f64x2, u64x2); -impl_shuffle1_dyn_non_u!(f64x4, u64x4); -impl_shuffle1_dyn_non_u!(f64x8, u64x8); - -// Implementation for non-unsigned vector types -macro_rules! impl_shuffle1_dyn_ptr { - ($id:ident, $uid:ident) => { - impl<T> Shuffle1Dyn for $id<T> { - type Indices = $uid; - #[inline] - fn shuffle1_dyn(self, indices: Self::Indices) -> Self { - unsafe { - let u: $uid = crate::mem::transmute(self); - crate::mem::transmute(u.shuffle1_dyn(indices)) - } - } - } - }; -} - -impl_shuffle1_dyn_ptr!(cptrx2, usizex2); -impl_shuffle1_dyn_ptr!(cptrx4, usizex4); -impl_shuffle1_dyn_ptr!(cptrx8, usizex8); - -impl_shuffle1_dyn_ptr!(mptrx2, usizex2); -impl_shuffle1_dyn_ptr!(mptrx4, usizex4); -impl_shuffle1_dyn_ptr!(mptrx8, usizex8); diff --git a/third_party/rust/packed_simd/src/codegen/swap_bytes.rs b/third_party/rust/packed_simd/src/codegen/swap_bytes.rs deleted file mode 100644 index 9cf34a3e04..0000000000 --- a/third_party/rust/packed_simd/src/codegen/swap_bytes.rs +++ /dev/null @@ -1,149 +0,0 @@ -//! Horizontal swap bytes reductions. - -// FIXME: investigate using `llvm.bswap` -// https://github.com/rust-lang-nursery/packed_simd/issues/19 - -use crate::*; - -pub(crate) trait SwapBytes { - fn swap_bytes(self) -> Self; -} - -macro_rules! impl_swap_bytes { - (v16: $($id:ident,)+) => { - $( - impl SwapBytes for $id { - #[inline] - fn swap_bytes(self) -> Self { - shuffle!(self, [1, 0]) - } - } - )+ - }; - (v32: $($id:ident,)+) => { - $( - impl SwapBytes for $id { - #[inline] - #[allow(clippy::useless_transmute)] - fn swap_bytes(self) -> Self { - unsafe { - let bytes: u8x4 = crate::mem::transmute(self); - let result: u8x4 = shuffle!(bytes, [3, 2, 1, 0]); - crate::mem::transmute(result) - } - } - } - )+ - }; - (v64: $($id:ident,)+) => { - $( - impl SwapBytes for $id { - #[inline] - #[allow(clippy::useless_transmute)] - fn swap_bytes(self) -> Self { - unsafe { - let bytes: u8x8 = crate::mem::transmute(self); - let result: u8x8 = shuffle!( - bytes, [7, 6, 5, 4, 3, 2, 1, 0] - ); - crate::mem::transmute(result) - } - } - } - )+ - }; - (v128: $($id:ident,)+) => { - $( - impl SwapBytes for $id { - #[inline] - #[allow(clippy::useless_transmute)] - fn swap_bytes(self) -> Self { - unsafe { - let bytes: u8x16 = crate::mem::transmute(self); - let result: u8x16 = shuffle!(bytes, [ - 15, 14, 13, 12, 11, 10, 9, 8, - 7, 6, 5, 4, 3, 2, 1, 0 - ]); - crate::mem::transmute(result) - } - } - } - )+ - }; - (v256: $($id:ident,)+) => { - $( - impl SwapBytes for $id { - #[inline] - #[allow(clippy::useless_transmute)] - fn swap_bytes(self) -> Self { - unsafe { - let bytes: u8x32 = crate::mem::transmute(self); - let result: u8x32 = shuffle!(bytes, [ - 31, 30, 29, 28, 27, 26, 25, 24, - 23, 22, 21, 20, 19, 18, 17, 16, - 15, 14, 13, 12, 11, 10, 9, 8, - 7, 6, 5, 4, 3, 2, 1, 0 - ]); - crate::mem::transmute(result) - } - } - } - )+ - }; - (v512: $($id:ident,)+) => { - $( - impl SwapBytes for $id { - #[inline] - #[allow(clippy::useless_transmute)] - fn swap_bytes(self) -> Self { - unsafe { - let bytes: u8x64 = crate::mem::transmute(self); - let result: u8x64 = shuffle!(bytes, [ - 63, 62, 61, 60, 59, 58, 57, 56, - 55, 54, 53, 52, 51, 50, 49, 48, - 47, 46, 45, 44, 43, 42, 41, 40, - 39, 38, 37, 36, 35, 34, 33, 32, - 31, 30, 29, 28, 27, 26, 25, 24, - 23, 22, 21, 20, 19, 18, 17, 16, - 15, 14, 13, 12, 11, 10, 9, 8, - 7, 6, 5, 4, 3, 2, 1, 0 - ]); - crate::mem::transmute(result) - } - } - } - )+ - }; -} - -impl_swap_bytes!(v16: u8x2, i8x2,); -impl_swap_bytes!(v32: u8x4, i8x4, u16x2, i16x2,); -// FIXME: 64-bit single element vector -impl_swap_bytes!(v64: u8x8, i8x8, u16x4, i16x4, u32x2, i32x2 /* u64x1, i64x1, */,); - -impl_swap_bytes!(v128: u8x16, i8x16, u16x8, i16x8, u32x4, i32x4, u64x2, i64x2, u128x1, i128x1,); -impl_swap_bytes!(v256: u8x32, i8x32, u16x16, i16x16, u32x8, i32x8, u64x4, i64x4, u128x2, i128x2,); - -impl_swap_bytes!(v512: u8x64, i8x64, u16x32, i16x32, u32x16, i32x16, u64x8, i64x8, u128x4, i128x4,); - -cfg_if! { - if #[cfg(target_pointer_width = "8")] { - impl_swap_bytes!(v16: isizex2, usizex2,); - impl_swap_bytes!(v32: isizex4, usizex4,); - impl_swap_bytes!(v64: isizex8, usizex8,); - } else if #[cfg(target_pointer_width = "16")] { - impl_swap_bytes!(v32: isizex2, usizex2,); - impl_swap_bytes!(v64: isizex4, usizex4,); - impl_swap_bytes!(v128: isizex8, usizex8,); - } else if #[cfg(target_pointer_width = "32")] { - impl_swap_bytes!(v64: isizex2, usizex2,); - impl_swap_bytes!(v128: isizex4, usizex4,); - impl_swap_bytes!(v256: isizex8, usizex8,); - } else if #[cfg(target_pointer_width = "64")] { - impl_swap_bytes!(v128: isizex2, usizex2,); - impl_swap_bytes!(v256: isizex4, usizex4,); - impl_swap_bytes!(v512: isizex8, usizex8,); - } else { - compile_error!("unsupported target_pointer_width"); - } -} diff --git a/third_party/rust/packed_simd/src/codegen/v128.rs b/third_party/rust/packed_simd/src/codegen/v128.rs deleted file mode 100644 index 9506424fad..0000000000 --- a/third_party/rust/packed_simd/src/codegen/v128.rs +++ /dev/null @@ -1,46 +0,0 @@ -//! Internal 128-bit wide vector types - -use crate::masks::*; - -#[rustfmt::skip] -impl_simd_array!( - [i8; 16]: i8x16 | - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8 -); -#[rustfmt::skip] -impl_simd_array!( - [u8; 16]: u8x16 | - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8 -); -#[rustfmt::skip] -impl_simd_array!( - [m8; 16]: m8x16 | - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8 -); - -impl_simd_array!([i16; 8]: i16x8 | i16, i16, i16, i16, i16, i16, i16, i16); -impl_simd_array!([u16; 8]: u16x8 | u16, u16, u16, u16, u16, u16, u16, u16); -impl_simd_array!([m16; 8]: m16x8 | i16, i16, i16, i16, i16, i16, i16, i16); - -impl_simd_array!([i32; 4]: i32x4 | i32, i32, i32, i32); -impl_simd_array!([u32; 4]: u32x4 | u32, u32, u32, u32); -impl_simd_array!([f32; 4]: f32x4 | f32, f32, f32, f32); -impl_simd_array!([m32; 4]: m32x4 | i32, i32, i32, i32); - -impl_simd_array!([i64; 2]: i64x2 | i64, i64); -impl_simd_array!([u64; 2]: u64x2 | u64, u64); -impl_simd_array!([f64; 2]: f64x2 | f64, f64); -impl_simd_array!([m64; 2]: m64x2 | i64, i64); - -impl_simd_array!([i128; 1]: i128x1 | i128); -impl_simd_array!([u128; 1]: u128x1 | u128); -impl_simd_array!([m128; 1]: m128x1 | i128); diff --git a/third_party/rust/packed_simd/src/codegen/v16.rs b/third_party/rust/packed_simd/src/codegen/v16.rs deleted file mode 100644 index 4d55a6d899..0000000000 --- a/third_party/rust/packed_simd/src/codegen/v16.rs +++ /dev/null @@ -1,7 +0,0 @@ -//! Internal 16-bit wide vector types - -use crate::masks::*; - -impl_simd_array!([i8; 2]: i8x2 | i8, i8); -impl_simd_array!([u8; 2]: u8x2 | u8, u8); -impl_simd_array!([m8; 2]: m8x2 | i8, i8); diff --git a/third_party/rust/packed_simd/src/codegen/v256.rs b/third_party/rust/packed_simd/src/codegen/v256.rs deleted file mode 100644 index 5ca4759f0c..0000000000 --- a/third_party/rust/packed_simd/src/codegen/v256.rs +++ /dev/null @@ -1,78 +0,0 @@ -//! Internal 256-bit wide vector types - -use crate::masks::*; - -#[rustfmt::skip] -impl_simd_array!( - [i8; 32]: i8x32 | - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8 -); -#[rustfmt::skip] -impl_simd_array!( - [u8; 32]: u8x32 | - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8 -); -#[rustfmt::skip] -impl_simd_array!( - [m8; 32]: m8x32 | - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8 -); -#[rustfmt::skip] -impl_simd_array!( - [i16; 16]: i16x16 | - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16 -); -#[rustfmt::skip] -impl_simd_array!( - [u16; 16]: u16x16 | - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16 -); -#[rustfmt::skip] -impl_simd_array!( - [m16; 16]: m16x16 | - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16 -); - -impl_simd_array!([i32; 8]: i32x8 | i32, i32, i32, i32, i32, i32, i32, i32); -impl_simd_array!([u32; 8]: u32x8 | u32, u32, u32, u32, u32, u32, u32, u32); -impl_simd_array!([f32; 8]: f32x8 | f32, f32, f32, f32, f32, f32, f32, f32); -impl_simd_array!([m32; 8]: m32x8 | i32, i32, i32, i32, i32, i32, i32, i32); - -impl_simd_array!([i64; 4]: i64x4 | i64, i64, i64, i64); -impl_simd_array!([u64; 4]: u64x4 | u64, u64, u64, u64); -impl_simd_array!([f64; 4]: f64x4 | f64, f64, f64, f64); -impl_simd_array!([m64; 4]: m64x4 | i64, i64, i64, i64); - -impl_simd_array!([i128; 2]: i128x2 | i128, i128); -impl_simd_array!([u128; 2]: u128x2 | u128, u128); -impl_simd_array!([m128; 2]: m128x2 | i128, i128); diff --git a/third_party/rust/packed_simd/src/codegen/v32.rs b/third_party/rust/packed_simd/src/codegen/v32.rs deleted file mode 100644 index ae1dabd00c..0000000000 --- a/third_party/rust/packed_simd/src/codegen/v32.rs +++ /dev/null @@ -1,11 +0,0 @@ -//! Internal 32-bit wide vector types - -use crate::masks::*; - -impl_simd_array!([i8; 4]: i8x4 | i8, i8, i8, i8); -impl_simd_array!([u8; 4]: u8x4 | u8, u8, u8, u8); -impl_simd_array!([m8; 4]: m8x4 | i8, i8, i8, i8); - -impl_simd_array!([i16; 2]: i16x2 | i16, i16); -impl_simd_array!([u16; 2]: u16x2 | u16, u16); -impl_simd_array!([m16; 2]: m16x2 | i16, i16); diff --git a/third_party/rust/packed_simd/src/codegen/v512.rs b/third_party/rust/packed_simd/src/codegen/v512.rs deleted file mode 100644 index bf95110340..0000000000 --- a/third_party/rust/packed_simd/src/codegen/v512.rs +++ /dev/null @@ -1,145 +0,0 @@ -//! Internal 512-bit wide vector types - -use crate::masks::*; - -#[rustfmt::skip] -impl_simd_array!( - [i8; 64]: i8x64 | - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8 -); -#[rustfmt::skip] -impl_simd_array!( - [u8; 64]: u8x64 | - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8, - u8, u8, u8, u8 -); -#[rustfmt::skip] -impl_simd_array!( - [m8; 64]: m8x64 | - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8, - i8, i8, i8, i8 -); -#[rustfmt::skip] -impl_simd_array!( - [i16; 32]: i16x32 | - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16 -); -#[rustfmt::skip] -impl_simd_array!( - [u16; 32]: u16x32 | - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16, - u16, u16, u16, u16 -); -#[rustfmt::skip] -impl_simd_array!( - [m16; 32]: m16x32 | - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16, - i16, i16, i16, i16 -); - -#[rustfmt::skip] -impl_simd_array!( - [i32; 16]: i32x16 | - i32, i32, i32, i32, - i32, i32, i32, i32, - i32, i32, i32, i32, - i32, i32, i32, i32 -); -#[rustfmt::skip] -impl_simd_array!( - [u32; 16]: u32x16 | - u32, u32, u32, u32, - u32, u32, u32, u32, - u32, u32, u32, u32, - u32, u32, u32, u32 -); -#[rustfmt::skip] -impl_simd_array!( - [f32; 16]: f32x16 | - f32, f32, f32, f32, - f32, f32, f32, f32, - f32, f32, f32, f32, - f32, f32, f32, f32 -); -#[rustfmt::skip] -impl_simd_array!( - [m32; 16]: m32x16 | - i32, i32, i32, i32, - i32, i32, i32, i32, - i32, i32, i32, i32, - i32, i32, i32, i32 -); - -impl_simd_array!([i64; 8]: i64x8 | i64, i64, i64, i64, i64, i64, i64, i64); -impl_simd_array!([u64; 8]: u64x8 | u64, u64, u64, u64, u64, u64, u64, u64); -impl_simd_array!([f64; 8]: f64x8 | f64, f64, f64, f64, f64, f64, f64, f64); -impl_simd_array!([m64; 8]: m64x8 | i64, i64, i64, i64, i64, i64, i64, i64); - -impl_simd_array!([i128; 4]: i128x4 | i128, i128, i128, i128); -impl_simd_array!([u128; 4]: u128x4 | u128, u128, u128, u128); -impl_simd_array!([m128; 4]: m128x4 | i128, i128, i128, i128); diff --git a/third_party/rust/packed_simd/src/codegen/v64.rs b/third_party/rust/packed_simd/src/codegen/v64.rs deleted file mode 100644 index 3cfb67c1a0..0000000000 --- a/third_party/rust/packed_simd/src/codegen/v64.rs +++ /dev/null @@ -1,21 +0,0 @@ -//! Internal 64-bit wide vector types - -use crate::masks::*; - -impl_simd_array!([i8; 8]: i8x8 | i8, i8, i8, i8, i8, i8, i8, i8); -impl_simd_array!([u8; 8]: u8x8 | u8, u8, u8, u8, u8, u8, u8, u8); -impl_simd_array!([m8; 8]: m8x8 | i8, i8, i8, i8, i8, i8, i8, i8); - -impl_simd_array!([i16; 4]: i16x4 | i16, i16, i16, i16); -impl_simd_array!([u16; 4]: u16x4 | u16, u16, u16, u16); -impl_simd_array!([m16; 4]: m16x4 | i16, i16, i16, i16); - -impl_simd_array!([i32; 2]: i32x2 | i32, i32); -impl_simd_array!([u32; 2]: u32x2 | u32, u32); -impl_simd_array!([f32; 2]: f32x2 | f32, f32); -impl_simd_array!([m32; 2]: m32x2 | i32, i32); - -impl_simd_array!([i64; 1]: i64x1 | i64); -impl_simd_array!([u64; 1]: u64x1 | u64); -impl_simd_array!([f64; 1]: f64x1 | f64); -impl_simd_array!([m64; 1]: m64x1 | i64); diff --git a/third_party/rust/packed_simd/src/codegen/vPtr.rs b/third_party/rust/packed_simd/src/codegen/vPtr.rs deleted file mode 100644 index abd3aa8779..0000000000 --- a/third_party/rust/packed_simd/src/codegen/vPtr.rs +++ /dev/null @@ -1,35 +0,0 @@ -//! Pointer vector types - -macro_rules! impl_simd_ptr { - ([$ptr_ty:ty; $elem_count:expr]: $tuple_id:ident | $ty:ident - | $($tys:ty),*) => { - #[derive(Copy, Clone)] - #[repr(simd)] - pub struct $tuple_id<$ty>($(pub(crate) $tys),*); - //^^^^^^^ leaked through SimdArray - - impl<$ty> crate::sealed::Seal for [$ptr_ty; $elem_count] {} - impl<$ty> crate::sealed::SimdArray for [$ptr_ty; $elem_count] { - type Tuple = $tuple_id<$ptr_ty>; - type T = $ptr_ty; - const N: usize = $elem_count; - type NT = [u32; $elem_count]; - } - - impl<$ty> crate::sealed::Seal for $tuple_id<$ptr_ty> {} - impl<$ty> crate::sealed::Simd for $tuple_id<$ptr_ty> { - type Element = $ptr_ty; - const LANES: usize = $elem_count; - type LanesType = [u32; $elem_count]; - } - - } -} - -impl_simd_ptr!([*const T; 2]: cptrx2 | T | T, T); -impl_simd_ptr!([*const T; 4]: cptrx4 | T | T, T, T, T); -impl_simd_ptr!([*const T; 8]: cptrx8 | T | T, T, T, T, T, T, T, T); - -impl_simd_ptr!([*mut T; 2]: mptrx2 | T | T, T); -impl_simd_ptr!([*mut T; 4]: mptrx4 | T | T, T, T, T); -impl_simd_ptr!([*mut T; 8]: mptrx8 | T | T, T, T, T, T, T, T, T); diff --git a/third_party/rust/packed_simd/src/codegen/vSize.rs b/third_party/rust/packed_simd/src/codegen/vSize.rs deleted file mode 100644 index d5db03991d..0000000000 --- a/third_party/rust/packed_simd/src/codegen/vSize.rs +++ /dev/null @@ -1,16 +0,0 @@ -//! Vector types with pointer-sized elements - -use crate::codegen::pointer_sized_int::{isize_, usize_}; -use crate::masks::*; - -impl_simd_array!([isize; 2]: isizex2 | isize_, isize_); -impl_simd_array!([usize; 2]: usizex2 | usize_, usize_); -impl_simd_array!([msize; 2]: msizex2 | isize_, isize_); - -impl_simd_array!([isize; 4]: isizex4 | isize_, isize_, isize_, isize_); -impl_simd_array!([usize; 4]: usizex4 | usize_, usize_, usize_, usize_); -impl_simd_array!([msize; 4]: msizex4 | isize_, isize_, isize_, isize_); - -impl_simd_array!([isize; 8]: isizex8 | isize_, isize_, isize_, isize_, isize_, isize_, isize_, isize_); -impl_simd_array!([usize; 8]: usizex8 | usize_, usize_, usize_, usize_, usize_, usize_, usize_, usize_); -impl_simd_array!([msize; 8]: msizex8 | isize_, isize_, isize_, isize_, isize_, isize_, isize_, isize_); diff --git a/third_party/rust/packed_simd/src/lib.rs b/third_party/rust/packed_simd/src/lib.rs deleted file mode 100644 index 867cc10e9e..0000000000 --- a/third_party/rust/packed_simd/src/lib.rs +++ /dev/null @@ -1,348 +0,0 @@ -//! # Portable packed SIMD vectors -//! -//! This crate is proposed for stabilization as `std::packed_simd` in [RFC2366: -//! `std::simd`](https://github.com/rust-lang/rfcs/pull/2366) . -//! -//! The examples available in the -//! [`examples/`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples) -//! sub-directory of the crate showcase how to use the library in practice. -//! -//! ## Table of contents -//! -//! - [Introduction](#introduction) -//! - [Vector types](#vector-types) -//! - [Conditional operations](#conditional-operations) -//! - [Conversions](#conversions) -//! - [Hardware Features](#hardware-features) -//! - [Performance guide](https://rust-lang-nursery.github.io/packed_simd/perf-guide/) -//! -//! ## Introduction -//! -//! This crate exports [`Simd<[T; N]>`][`Simd`]: a packed vector of `N` -//! elements of type `T` as well as many type aliases for this type: for -//! example, [`f32x4`], which is just an alias for `Simd<[f32; 4]>`. -//! -//! The operations on packed vectors are, by default, "vertical", that is, they -//! are applied to each vector lane in isolation of the others: -//! -//! ``` -//! # use packed_simd::*; -//! let a = i32x4::new(1, 2, 3, 4); -//! let b = i32x4::new(5, 6, 7, 8); -//! assert_eq!(a + b, i32x4::new(6, 8, 10, 12)); -//! ``` -//! -//! Many "horizontal" operations are also provided: -//! -//! ``` -//! # use packed_simd::*; -//! # let a = i32x4::new(1, 2, 3, 4); -//! assert_eq!(a.wrapping_sum(), 10); -//! ``` -//! -//! In virtually all architectures vertical operations are fast, while -//! horizontal operations are, by comparison, much slower. That is, the -//! most portably-efficient way of performing a reduction over a slice -//! is to collect the results into a vector using vertical operations, -//! and performing a single horizontal operation at the end: -//! -//! ``` -//! # use packed_simd::*; -//! fn reduce(x: &[i32]) -> i32 { -//! assert_eq!(x.len() % 4, 0); -//! let mut sum = i32x4::splat(0); // [0, 0, 0, 0] -//! for i in (0..x.len()).step_by(4) { -//! sum += i32x4::from_slice_unaligned(&x[i..]); -//! } -//! sum.wrapping_sum() -//! } -//! -//! let x = [0, 1, 2, 3, 4, 5, 6, 7]; -//! assert_eq!(reduce(&x), 28); -//! ``` -//! -//! ## Vector types -//! -//! The vector type aliases are named according to the following scheme: -//! -//! > `{element_type}x{number_of_lanes} == Simd<[element_type; -//! number_of_lanes]>` -//! -//! where the following element types are supported: -//! -//! * `i{element_width}`: signed integer -//! * `u{element_width}`: unsigned integer -//! * `f{element_width}`: float -//! * `m{element_width}`: mask (see below) -//! * `*{const,mut} T`: `const` and `mut` pointers -//! -//! ## Basic operations -//! -//! ``` -//! # use packed_simd::*; -//! // Sets all elements to `0`: -//! let a = i32x4::splat(0); -//! -//! // Reads a vector from a slice: -//! let mut arr = [0, 0, 0, 1, 2, 3, 4, 5]; -//! let b = i32x4::from_slice_unaligned(&arr); -//! -//! // Reads the 4-th element of a vector: -//! assert_eq!(b.extract(3), 1); -//! -//! // Returns a new vector where the 4-th element is replaced with `1`: -//! let a = a.replace(3, 1); -//! assert_eq!(a, b); -//! -//! // Writes a vector to a slice: -//! let a = a.replace(2, 1); -//! a.write_to_slice_unaligned(&mut arr[4..]); -//! assert_eq!(arr, [0, 0, 0, 1, 0, 0, 1, 1]); -//! ``` -//! -//! ## Conditional operations -//! -//! One often needs to perform an operation on some lanes of the vector. Vector -//! masks, like `m32x4`, allow selecting on which vector lanes an operation is -//! to be performed: -//! -//! ``` -//! # use packed_simd::*; -//! let a = i32x4::new(1, 1, 2, 2); -//! -//! // Add `1` to the first two lanes of the vector. -//! let m = m16x4::new(true, true, false, false); -//! let a = m.select(a + 1, a); -//! assert_eq!(a, i32x4::splat(2)); -//! ``` -//! -//! The elements of a vector mask are either `true` or `false`. Here `true` -//! means that a lane is "selected", while `false` means that a lane is not -//! selected. -//! -//! All vector masks implement a `mask.select(a: T, b: T) -> T` method that -//! works on all vectors that have the same number of lanes as the mask. The -//! resulting vector contains the elements of `a` for those lanes for which the -//! mask is `true`, and the elements of `b` otherwise. -//! -//! The example constructs a mask with the first two lanes set to `true` and -//! the last two lanes set to `false`. This selects the first two lanes of `a + -//! 1` and the last two lanes of `a`, producing a vector where the first two -//! lanes have been incremented by `1`. -//! -//! > note: mask `select` can be used on vector types that have the same number -//! > of lanes as the mask. The example shows this by using [`m16x4`] instead -//! > of [`m32x4`]. It is _typically_ more performant to use a mask element -//! > width equal to the element width of the vectors being operated upon. -//! > This is, however, not true for 512-bit wide vectors when targeting -//! > AVX-512, where the most efficient masks use only 1-bit per element. -//! -//! All vertical comparison operations returns masks: -//! -//! ``` -//! # use packed_simd::*; -//! let a = i32x4::new(1, 1, 3, 3); -//! let b = i32x4::new(2, 2, 0, 0); -//! -//! // ge: >= (Greater Eequal; see also lt, le, gt, eq, ne). -//! let m = a.ge(i32x4::splat(2)); -//! -//! if m.any() { -//! // all / any / none allow coherent control flow -//! let d = m.select(a, b); -//! assert_eq!(d, i32x4::new(2, 2, 3, 3)); -//! } -//! ``` -//! -//! ## Conversions -//! -//! * **lossless widening conversions**: [`From`]/[`Into`] are implemented for -//! vectors with the same number of lanes when the conversion is value -//! preserving (same as in `std`). -//! -//! * **safe bitwise conversions**: The cargo feature `into_bits` provides the -//! `IntoBits/FromBits` traits (`x.into_bits()`). These perform safe bitwise -//! `transmute`s when all bit patterns of the source type are valid bit -//! patterns of the target type and are also implemented for the -//! architecture-specific vector types of `std::arch`. For example, `let x: -//! u8x8 = m8x8::splat(true).into_bits();` is provided because all `m8x8` bit -//! patterns are valid `u8x8` bit patterns. However, the opposite is not -//! true, not all `u8x8` bit patterns are valid `m8x8` bit-patterns, so this -//! operation cannot be performed safely using `x.into_bits()`; one needs to -//! use `unsafe { crate::mem::transmute(x) }` for that, making sure that the -//! value in the `u8x8` is a valid bit-pattern of `m8x8`. -//! -//! * **numeric casts** (`as`): are performed using [`FromCast`]/[`Cast`] -//! (`x.cast()`), just like `as`: -//! -//! * casting integer vectors whose lane types have the same size (e.g. -//! `i32xN` -> `u32xN`) is a **no-op**, -//! -//! * casting from a larger integer to a smaller integer (e.g. `u32xN` -> -//! `u8xN`) will **truncate**, -//! -//! * casting from a smaller integer to a larger integer (e.g. `u8xN` -> -//! `u32xN`) will: -//! * **zero-extend** if the source is unsigned, or -//! * **sign-extend** if the source is signed, -//! -//! * casting from a float to an integer will **round the float towards -//! zero**, -//! -//! * casting from an integer to float will produce the floating point -//! representation of the integer, **rounding to nearest, ties to even**, -//! -//! * casting from an `f32` to an `f64` is perfect and lossless, -//! -//! * casting from an `f64` to an `f32` **rounds to nearest, ties to even**. -//! -//! Numeric casts are not very "precise": sometimes lossy, sometimes value -//! preserving, etc. -//! -//! ## Hardware Features -//! -//! This crate can use different hardware features based on your configured -//! `RUSTFLAGS`. For example, with no configured `RUSTFLAGS`, `u64x8` on -//! x86_64 will use SSE2 operations like `PCMPEQD`. If you configure -//! `RUSTFLAGS='-C target-feature=+avx2,+avx'` on supported x86_64 hardware -//! the same `u64x8` may use wider AVX2 operations like `VPCMPEQQ`. It is -//! important for performance and for hardware support requirements that -//! you choose an appropriate set of `target-feature` and `target-cpu` -//! options during builds. For more information, see the [Performance -//! guide](https://rust-lang-nursery.github.io/packed_simd/perf-guide/) - -#![feature( - adt_const_params, - repr_simd, - rustc_attrs, - platform_intrinsics, - stdsimd, - arm_target_feature, - link_llvm_intrinsics, - core_intrinsics, - stmt_expr_attributes, - custom_inner_attributes, -)] -#![allow(non_camel_case_types, non_snake_case, - // FIXME: these types are unsound in C FFI already - // See https://github.com/rust-lang/rust/issues/53346 - improper_ctypes_definitions, - incomplete_features, - clippy::cast_possible_truncation, - clippy::cast_lossless, - clippy::cast_possible_wrap, - clippy::cast_precision_loss, - // TODO: manually add the `#[must_use]` attribute where appropriate - clippy::must_use_candidate, - // This lint is currently broken for generic code - // See https://github.com/rust-lang/rust-clippy/issues/3410 - clippy::use_self, - clippy::wrong_self_convention, - clippy::from_over_into, -)] -#![cfg_attr(test, feature(hashmap_internals))] -#![cfg_attr(doc_cfg, feature(doc_cfg))] -#![deny(rust_2018_idioms, clippy::missing_inline_in_public_items)] -#![no_std] - -use cfg_if::cfg_if; - -cfg_if! { - if #[cfg(feature = "core_arch")] { - #[allow(unused_imports)] - use core_arch as arch; - } else { - #[allow(unused_imports)] - use core::arch; - } -} - -#[cfg(all(target_arch = "wasm32", test))] -use wasm_bindgen_test::*; - -#[allow(unused_imports)] -use core::{ - /* arch (handled above), */ cmp, f32, f64, fmt, hash, hint, i128, i16, i32, i64, i8, intrinsics, - isize, iter, marker, mem, ops, ptr, slice, u128, u16, u32, u64, u8, usize, -}; - -#[macro_use] -mod testing; -#[macro_use] -mod api; -mod codegen; -mod sealed; - -pub use crate::sealed::{Mask, Shuffle, Simd as SimdVector, SimdArray}; - -/// Packed SIMD vector type. -/// -/// # Examples -/// -/// ``` -/// # use packed_simd::Simd; -/// let v = Simd::<[i32; 4]>::new(0, 1, 2, 3); -/// assert_eq!(v.extract(2), 2); -/// ``` -#[repr(transparent)] -#[derive(Copy, Clone)] -pub struct Simd<A: sealed::SimdArray>( - // FIXME: this type should be private, - // but it currently must be public for the - // `shuffle!` macro to work: it needs to - // access the internal `repr(simd)` type - // to call the shuffle intrinsics. - #[doc(hidden)] pub <A as sealed::SimdArray>::Tuple, -); - -impl<A: sealed::SimdArray> sealed::Seal for Simd<A> {} - -/// Wrapper over `T` implementing a lexicoraphical order via the `PartialOrd` -/// and/or `Ord` traits. -#[repr(transparent)] -#[derive(Copy, Clone, Debug)] -#[allow(clippy::missing_inline_in_public_items)] -pub struct LexicographicallyOrdered<T>(T); - -mod masks; -pub use self::masks::*; - -mod v16; -pub use self::v16::*; - -mod v32; -pub use self::v32::*; - -mod v64; -pub use self::v64::*; - -mod v128; -pub use self::v128::*; - -mod v256; -pub use self::v256::*; - -mod v512; -pub use self::v512::*; - -mod vSize; -pub use self::vSize::*; - -mod vPtr; -pub use self::vPtr::*; - -pub use self::api::cast::*; - -#[cfg(feature = "into_bits")] -pub use self::api::into_bits::*; - -// Re-export the shuffle intrinsics required by the `shuffle!` macro. -#[doc(hidden)] -pub use self::codegen::llvm::{ - __shuffle_vector16, __shuffle_vector2, __shuffle_vector32, __shuffle_vector4, __shuffle_vector64, - __shuffle_vector8, -}; - -pub(crate) mod llvm { - pub(crate) use crate::codegen::llvm::*; -} diff --git a/third_party/rust/packed_simd/src/masks.rs b/third_party/rust/packed_simd/src/masks.rs deleted file mode 100644 index 04534eab2b..0000000000 --- a/third_party/rust/packed_simd/src/masks.rs +++ /dev/null @@ -1,126 +0,0 @@ -//! Mask types - -macro_rules! impl_mask_ty { - ($id:ident : $elem_ty:ident | #[$doc:meta]) => { - #[$doc] - #[derive(Copy, Clone)] - pub struct $id($elem_ty); - - impl crate::sealed::Seal for $id {} - impl crate::sealed::Mask for $id { - #[inline] - fn test(&self) -> bool { - $id::test(self) - } - } - - impl $id { - /// Instantiate a mask with `value` - #[inline] - pub fn new(x: bool) -> Self { - if x { - $id(!0) - } else { - $id(0) - } - } - /// Test if the mask is set - #[inline] - pub fn test(&self) -> bool { - self.0 != 0 - } - } - - impl Default for $id { - #[inline] - fn default() -> Self { - $id(0) - } - } - - #[allow(clippy::partialeq_ne_impl)] - impl PartialEq<$id> for $id { - #[inline] - fn eq(&self, other: &Self) -> bool { - self.0 == other.0 - } - #[inline] - fn ne(&self, other: &Self) -> bool { - self.0 != other.0 - } - } - - impl Eq for $id {} - - impl PartialOrd<$id> for $id { - #[inline] - fn partial_cmp(&self, other: &Self) -> Option<crate::cmp::Ordering> { - use crate::cmp::Ordering; - if self == other { - Some(Ordering::Equal) - } else if self.0 > other.0 { - // Note: - // * false = 0_i - // * true == !0_i == -1_i - Some(Ordering::Less) - } else { - Some(Ordering::Greater) - } - } - - #[inline] - fn lt(&self, other: &Self) -> bool { - self.0 > other.0 - } - #[inline] - fn gt(&self, other: &Self) -> bool { - self.0 < other.0 - } - #[inline] - fn le(&self, other: &Self) -> bool { - self.0 >= other.0 - } - #[inline] - fn ge(&self, other: &Self) -> bool { - self.0 <= other.0 - } - } - - impl Ord for $id { - #[inline] - fn cmp(&self, other: &Self) -> crate::cmp::Ordering { - match self.partial_cmp(other) { - Some(x) => x, - None => unsafe { crate::hint::unreachable_unchecked() }, - } - } - } - - impl crate::hash::Hash for $id { - #[inline] - fn hash<H: crate::hash::Hasher>(&self, state: &mut H) { - (self.0 != 0).hash(state); - } - } - - impl crate::fmt::Debug for $id { - #[inline] - fn fmt(&self, fmtter: &mut crate::fmt::Formatter<'_>) -> Result<(), crate::fmt::Error> { - write!(fmtter, "{}({})", stringify!($id), self.0 != 0) - } - } - }; -} - -impl_mask_ty!(m8: i8 | /// 8-bit wide mask. -); -impl_mask_ty!(m16: i16 | /// 16-bit wide mask. -); -impl_mask_ty!(m32: i32 | /// 32-bit wide mask. -); -impl_mask_ty!(m64: i64 | /// 64-bit wide mask. -); -impl_mask_ty!(m128: i128 | /// 128-bit wide mask. -); -impl_mask_ty!(msize: isize | /// isize-wide mask. -); diff --git a/third_party/rust/packed_simd/src/sealed.rs b/third_party/rust/packed_simd/src/sealed.rs deleted file mode 100644 index 0ec20300fd..0000000000 --- a/third_party/rust/packed_simd/src/sealed.rs +++ /dev/null @@ -1,42 +0,0 @@ -//! Sealed traits - -/// A sealed trait, this is logically private to the crate -/// and will prevent implementations from outside the crate -pub trait Seal<T = ()> {} - -/// Trait implemented by arrays that can be SIMD types. -pub trait SimdArray: Seal { - /// The type of the #[repr(simd)] type. - type Tuple: Copy + Clone; - /// The element type of the vector. - type T; - /// The number of elements in the array. - const N: usize; - /// The type: `[u32; Self::N]`. - type NT; -} - -/// This traits is used to constraint the arguments -/// and result type of the portable shuffles. -#[doc(hidden)] -pub trait Shuffle<Lanes>: Seal<Lanes> { - // Lanes is a `[u32; N]` where `N` is the number of vector lanes - - /// The result type of the shuffle. - type Output; -} - -/// This trait is implemented by all SIMD vector types. -pub trait Simd: Seal { - /// Element type of the SIMD vector - type Element; - /// The number of elements in the SIMD vector. - const LANES: usize; - /// The type: `[u32; Self::N]`. - type LanesType; -} - -/// This trait is implemented by all mask types -pub trait Mask: Seal { - fn test(&self) -> bool; -} diff --git a/third_party/rust/packed_simd/src/testing.rs b/third_party/rust/packed_simd/src/testing.rs deleted file mode 100644 index 6320b28055..0000000000 --- a/third_party/rust/packed_simd/src/testing.rs +++ /dev/null @@ -1,8 +0,0 @@ -//! Testing macros and other utilities. - -#[macro_use] -mod macros; - -#[cfg(test)] -#[macro_use] -pub(crate) mod utils; diff --git a/third_party/rust/packed_simd/src/testing/macros.rs b/third_party/rust/packed_simd/src/testing/macros.rs deleted file mode 100644 index 7bc4268b90..0000000000 --- a/third_party/rust/packed_simd/src/testing/macros.rs +++ /dev/null @@ -1,44 +0,0 @@ -//! Testing macros - -macro_rules! test_if { - ($cfg_tt:tt: $it:item) => { - #[cfg(any( - // Test everything if: - // - // * tests are enabled, - // * no features about exclusively testing - // specific vector classes are enabled - all(test, not(any( - test_v16, - test_v32, - test_v64, - test_v128, - test_v256, - test_v512, - test_none, // disables all tests - ))), - // Test if: - // - // * tests are enabled - // * a particular cfg token tree returns true - all(test, $cfg_tt), - ))] - $it - }; -} - -#[cfg(test)] -#[allow(unused)] -macro_rules! ref_ { - ($anything:tt) => { - &$anything - }; -} - -#[cfg(test)] -#[allow(unused)] -macro_rules! ref_mut_ { - ($anything:tt) => { - &mut $anything - }; -} diff --git a/third_party/rust/packed_simd/src/testing/utils.rs b/third_party/rust/packed_simd/src/testing/utils.rs deleted file mode 100644 index 7d8f395739..0000000000 --- a/third_party/rust/packed_simd/src/testing/utils.rs +++ /dev/null @@ -1,130 +0,0 @@ -//! Testing utilities - -#![allow(dead_code)] -// FIXME: Or don't. But it's true this is a problematic comparison. -#![allow(clippy::neg_cmp_op_on_partial_ord)] - -use crate::{cmp::PartialOrd, fmt::Debug, LexicographicallyOrdered}; - -/// Tests PartialOrd for `a` and `b` where `a < b` is true. -pub fn test_lt<T>(a: LexicographicallyOrdered<T>, b: LexicographicallyOrdered<T>) -where - LexicographicallyOrdered<T>: Debug + PartialOrd, -{ - assert!(a < b, "{:?}, {:?}", a, b); - assert!(b > a, "{:?}, {:?}", a, b); - - assert!(!(a == b), "{:?}, {:?}", a, b); - assert_ne!(a, b, "{:?}, {:?}", a, b); - - assert!(a <= b, "{:?}, {:?}", a, b); - assert!(b >= a, "{:?}, {:?}", a, b); - - // The elegance of the mathematical expression of irreflexivity is more - // than clippy can handle. - #[allow(clippy::eq_op)] - { - // Irreflexivity - assert!(!(a < a), "{:?}, {:?}", a, b); - assert!(!(b < b), "{:?}, {:?}", a, b); - assert!(!(a > a), "{:?}, {:?}", a, b); - assert!(!(b > b), "{:?}, {:?}", a, b); - - assert!(a <= a, "{:?}, {:?}", a, b); - assert!(b <= b, "{:?}, {:?}", a, b); - } -} - -/// Tests PartialOrd for `a` and `b` where `a <= b` is true. -pub fn test_le<T>(a: LexicographicallyOrdered<T>, b: LexicographicallyOrdered<T>) -where - LexicographicallyOrdered<T>: Debug + PartialOrd, -{ - assert!(a <= b, "{:?}, {:?}", a, b); - assert!(b >= a, "{:?}, {:?}", a, b); - - assert!(a <= b, "{:?}, {:?}", a, b); - assert!(b >= a, "{:?}, {:?}", a, b); - - if a == b { - assert!(!(a < b), "{:?}, {:?}", a, b); - assert!(!(b > a), "{:?}, {:?}", a, b); - - assert!(!(a != b), "{:?}, {:?}", a, b); - } else { - assert_ne!(a, b, "{:?}, {:?}", a, b); - test_lt(a, b); - } -} - -/// Test PartialOrd::partial_cmp for `a` and `b` returning `Ordering` -pub fn test_cmp<T>( - a: LexicographicallyOrdered<T>, - b: LexicographicallyOrdered<T>, - o: Option<crate::cmp::Ordering>, -) where - LexicographicallyOrdered<T>: PartialOrd + Debug, - T: Debug + crate::sealed::Simd + Copy + Clone, - <T as crate::sealed::Simd>::Element: Default + Copy + Clone + PartialOrd, -{ - assert!(T::LANES <= 64, "array length in these two arrays needs updating"); - let mut arr_a: [T::Element; 64] = [Default::default(); 64]; - let mut arr_b: [T::Element; 64] = [Default::default(); 64]; - - unsafe { crate::ptr::write_unaligned(arr_a.as_mut_ptr() as *mut LexicographicallyOrdered<T>, a) } - unsafe { crate::ptr::write_unaligned(arr_b.as_mut_ptr() as *mut LexicographicallyOrdered<T>, b) } - let expected = arr_a[0..T::LANES].partial_cmp(&arr_b[0..T::LANES]); - let result = a.partial_cmp(&b); - assert_eq!(expected, result, "{:?}, {:?}", a, b); - assert_eq!(o, result, "{:?}, {:?}", a, b); - match o { - Some(crate::cmp::Ordering::Less) => { - test_lt(a, b); - test_le(a, b); - } - Some(crate::cmp::Ordering::Greater) => { - test_lt(b, a); - test_le(b, a); - } - Some(crate::cmp::Ordering::Equal) => { - assert!(a == b, "{:?}, {:?}", a, b); - assert!(!(a != b), "{:?}, {:?}", a, b); - assert!(!(a < b), "{:?}, {:?}", a, b); - assert!(!(b < a), "{:?}, {:?}", a, b); - assert!(!(a > b), "{:?}, {:?}", a, b); - assert!(!(b > a), "{:?}, {:?}", a, b); - - test_le(a, b); - test_le(b, a); - } - None => { - assert!(!(a == b), "{:?}, {:?}", a, b); - assert!(!(a != b), "{:?}, {:?}", a, b); - assert!(!(a < b), "{:?}, {:?}", a, b); - assert!(!(a > b), "{:?}, {:?}", a, b); - assert!(!(b < a), "{:?}, {:?}", a, b); - assert!(!(b > a), "{:?}, {:?}", a, b); - assert!(!(a <= b), "{:?}, {:?}", a, b); - assert!(!(b <= a), "{:?}, {:?}", a, b); - assert!(!(a >= b), "{:?}, {:?}", a, b); - assert!(!(b >= a), "{:?}, {:?}", a, b); - } - } -} - -// Returns a tuple containing two distinct pointer values of the same type as -// the element type of the Simd vector `$id`. -#[allow(unused)] -macro_rules! ptr_vals { - ($id:ty) => { - // expands to an expression - #[allow(unused_unsafe)] - unsafe { - // all bits cleared - let clear: <$id as sealed::Simd>::Element = crate::mem::zeroed(); - // all bits set - let set: <$id as sealed::Simd>::Element = crate::mem::transmute(-1_isize); - (clear, set) - } - }; -} diff --git a/third_party/rust/packed_simd/src/v128.rs b/third_party/rust/packed_simd/src/v128.rs deleted file mode 100644 index 7949f6619a..0000000000 --- a/third_party/rust/packed_simd/src/v128.rs +++ /dev/null @@ -1,80 +0,0 @@ -//! 128-bit wide vector types -#[rustfmt::skip] - -use crate::*; - -impl_i!([i8; 16]: i8x16, m8x16 | i8, u16 | test_v128 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: | - /// A 128-bit vector with 16 `i8` lanes. -); -impl_u!([u8; 16]: u8x16, m8x16 | u8, u16 | test_v128 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: | - /// A 128-bit vector with 16 `u8` lanes. -); -impl_m!([m8; 16]: m8x16 | i8, u16 | test_v128 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: m16x16 | - /// A 128-bit vector mask with 16 `m8` lanes. -); - -impl_i!([i16; 8]: i16x8, m16x8 | i16, u8 | test_v128 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: i8x8, u8x8 | - /// A 128-bit vector with 8 `i16` lanes. -); -impl_u!([u16; 8]: u16x8, m16x8 | u16, u8 | test_v128 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: u8x8 | - /// A 128-bit vector with 8 `u16` lanes. -); -impl_m!([m16; 8]: m16x8 | i16, u8 | test_v128 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: m8x8, m32x8 | - /// A 128-bit vector mask with 8 `m16` lanes. -); - -impl_i!([i32; 4]: i32x4, m32x4 | i32, u8 | test_v128 | x0, x1, x2, x3 | - From: i8x4, u8x4, i16x4, u16x4 | - /// A 128-bit vector with 4 `i32` lanes. -); -impl_u!([u32; 4]: u32x4, m32x4 | u32, u8 | test_v128 | x0, x1, x2, x3 | - From: u8x4, u16x4 | - /// A 128-bit vector with 4 `u32` lanes. -); -impl_f!([f32; 4]: f32x4, m32x4 | f32 | test_v128 | x0, x1, x2, x3 | - From: i8x4, u8x4, i16x4, u16x4 | - /// A 128-bit vector with 4 `f32` lanes. -); -impl_m!([m32; 4]: m32x4 | i32, u8 | test_v128 | x0, x1, x2, x3 | - From: m8x4, m16x4, m64x4 | - /// A 128-bit vector mask with 4 `m32` lanes. -); - -impl_i!([i64; 2]: i64x2, m64x2 | i64, u8 | test_v128 | x0, x1 | - From: i8x2, u8x2, i16x2, u16x2, i32x2, u32x2 | - /// A 128-bit vector with 2 `i64` lanes. -); -impl_u!([u64; 2]: u64x2, m64x2 | u64, u8 | test_v128 | x0, x1 | - From: u8x2, u16x2, u32x2 | - /// A 128-bit vector with 2 `u64` lanes. -); -impl_f!([f64; 2]: f64x2, m64x2 | f64 | test_v128 | x0, x1 | - From: i8x2, u8x2, i16x2, u16x2, i32x2, u32x2, f32x2 | - /// A 128-bit vector with 2 `f64` lanes. -); -impl_m!([m64; 2]: m64x2 | i64, u8 | test_v128 | x0, x1 | - From: m8x2, m16x2, m32x2, m128x2 | - /// A 128-bit vector mask with 2 `m64` lanes. -); - -impl_i!([i128; 1]: i128x1, m128x1 | i128, u8 | test_v128 | x0 | - From: /*i8x1, u8x1, i16x1, u16x1, i32x1, u32x1, i64x1, u64x1 */ | // FIXME: unary small vector types - /// A 128-bit vector with 1 `i128` lane. -); -impl_u!([u128; 1]: u128x1, m128x1 | u128, u8 | test_v128 | x0 | - From: /*u8x1, u16x1, u32x1, u64x1 */ | // FIXME: unary small vector types - /// A 128-bit vector with 1 `u128` lane. -); -impl_m!([m128; 1]: m128x1 | i128, u8 | test_v128 | x0 | - From: /*m8x1, m16x1, m32x1, m64x1 */ | // FIXME: unary small vector types - /// A 128-bit vector mask with 1 `m128` lane. -); diff --git a/third_party/rust/packed_simd/src/v16.rs b/third_party/rust/packed_simd/src/v16.rs deleted file mode 100644 index 4ca5afb2a7..0000000000 --- a/third_party/rust/packed_simd/src/v16.rs +++ /dev/null @@ -1,16 +0,0 @@ -//! 16-bit wide vector types - -use crate::*; - -impl_i!([i8; 2]: i8x2, m8x2 | i8, u8 | test_v16 | x0, x1 | - From: | - /// A 16-bit vector with 2 `i8` lanes. -); -impl_u!([u8; 2]: u8x2, m8x2 | u8, u8 | test_v16 | x0, x1 | - From: | - /// A 16-bit vector with 2 `u8` lanes. -); -impl_m!([m8; 2]: m8x2 | i8, u8 | test_v16 | x0, x1 | - From: m16x2, m32x2, m64x2, m128x2 | - /// A 16-bit vector mask with 2 `m8` lanes. -); diff --git a/third_party/rust/packed_simd/src/v256.rs b/third_party/rust/packed_simd/src/v256.rs deleted file mode 100644 index f0c3bc2813..0000000000 --- a/third_party/rust/packed_simd/src/v256.rs +++ /dev/null @@ -1,86 +0,0 @@ -//! 256-bit wide vector types -#[rustfmt::skip] - -use crate::*; - -impl_i!([i8; 32]: i8x32, m8x32 | i8, u32 | test_v256 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | - From: | - /// A 256-bit vector with 32 `i8` lanes. -); -impl_u!([u8; 32]: u8x32, m8x32 | u8, u32 | test_v256 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | - From: | - /// A 256-bit vector with 32 `u8` lanes. -); -impl_m!([m8; 32]: m8x32 | i8, u32 | test_v256 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | - From: | - /// A 256-bit vector mask with 32 `m8` lanes. -); - -impl_i!([i16; 16]: i16x16, m16x16 | i16, u16 | test_v256 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: i8x16, u8x16 | - /// A 256-bit vector with 16 `i16` lanes. -); -impl_u!([u16; 16]: u16x16, m16x16 | u16, u16 | test_v256 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: u8x16 | - /// A 256-bit vector with 16 `u16` lanes. -); -impl_m!([m16; 16]: m16x16 | i16, u16 | test_v256 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: m8x16 | - /// A 256-bit vector mask with 16 `m16` lanes. -); - -impl_i!([i32; 8]: i32x8, m32x8 | i32, u8 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: i8x8, u8x8, i16x8, u16x8 | - /// A 256-bit vector with 8 `i32` lanes. -); -impl_u!([u32; 8]: u32x8, m32x8 | u32, u8 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: u8x8, u16x8 | - /// A 256-bit vector with 8 `u32` lanes. -); -impl_f!([f32; 8]: f32x8, m32x8 | f32 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: i8x8, u8x8, i16x8, u16x8 | - /// A 256-bit vector with 8 `f32` lanes. -); -impl_m!([m32; 8]: m32x8 | i32, u8 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: m8x8, m16x8 | - /// A 256-bit vector mask with 8 `m32` lanes. -); - -impl_i!([i64; 4]: i64x4, m64x4 | i64, u8 | test_v256 | x0, x1, x2, x3 | - From: i8x4, u8x4, i16x4, u16x4, i32x4, u32x4 | - /// A 256-bit vector with 4 `i64` lanes. -); -impl_u!([u64; 4]: u64x4, m64x4 | u64, u8 | test_v256 | x0, x1, x2, x3 | - From: u8x4, u16x4, u32x4 | - /// A 256-bit vector with 4 `u64` lanes. -); -impl_f!([f64; 4]: f64x4, m64x4 | f64 | test_v256 | x0, x1, x2, x3 | - From: i8x4, u8x4, i16x4, u16x4, i32x4, u32x4, f32x4 | - /// A 256-bit vector with 4 `f64` lanes. -); -impl_m!([m64; 4]: m64x4 | i64, u8 | test_v256 | x0, x1, x2, x3 | - From: m8x4, m16x4, m32x4 | - /// A 256-bit vector mask with 4 `m64` lanes. -); - -impl_i!([i128; 2]: i128x2, m128x2 | i128, u8 | test_v256 | x0, x1 | - From: i8x2, u8x2, i16x2, u16x2, i32x2, u32x2, i64x2, u64x2 | - /// A 256-bit vector with 2 `i128` lanes. -); -impl_u!([u128; 2]: u128x2, m128x2 | u128, u8 | test_v256 | x0, x1 | - From: u8x2, u16x2, u32x2, u64x2 | - /// A 256-bit vector with 2 `u128` lanes. -); -impl_m!([m128; 2]: m128x2 | i128, u8 | test_v256 | x0, x1 | - From: m8x2, m16x2, m32x2, m64x2 | - /// A 256-bit vector mask with 2 `m128` lanes. -); diff --git a/third_party/rust/packed_simd/src/v32.rs b/third_party/rust/packed_simd/src/v32.rs deleted file mode 100644 index 75a1838e5e..0000000000 --- a/third_party/rust/packed_simd/src/v32.rs +++ /dev/null @@ -1,29 +0,0 @@ -//! 32-bit wide vector types - -use crate::*; - -impl_i!([i8; 4]: i8x4, m8x4 | i8, u8 | test_v32 | x0, x1, x2, x3 | - From: | - /// A 32-bit vector with 4 `i8` lanes. -); -impl_u!([u8; 4]: u8x4, m8x4 | u8, u8 | test_v32 | x0, x1, x2, x3 | - From: | - /// A 32-bit vector with 4 `u8` lanes. -); -impl_m!([m8; 4]: m8x4 | i8, u8 | test_v32 | x0, x1, x2, x3 | - From: m16x4, m32x4, m64x4 | - /// A 32-bit vector mask with 4 `m8` lanes. -); - -impl_i!([i16; 2]: i16x2, m16x2 | i16, u8 | test_v32 | x0, x1 | - From: i8x2, u8x2 | - /// A 32-bit vector with 2 `i16` lanes. -); -impl_u!([u16; 2]: u16x2, m16x2 | u16, u8 | test_v32 | x0, x1 | - From: u8x2 | - /// A 32-bit vector with 2 `u16` lanes. -); -impl_m!([m16; 2]: m16x2 | i16, u8 | test_v32 | x0, x1 | - From: m8x2, m32x2, m64x2, m128x2 | - /// A 32-bit vector mask with 2 `m16` lanes. -); diff --git a/third_party/rust/packed_simd/src/v512.rs b/third_party/rust/packed_simd/src/v512.rs deleted file mode 100644 index 4c8c71338a..0000000000 --- a/third_party/rust/packed_simd/src/v512.rs +++ /dev/null @@ -1,99 +0,0 @@ -//! 512-bit wide vector types -#[rustfmt::skip] - -use crate::*; - -impl_i!([i8; 64]: i8x64, m8x64 | i8, u64 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31, - x32, x33, x34, x35, x36, x37, x38, x39, x40, x41, x42, x43, x44, x45, x46, x47, - x48, x49, x50, x51, x52, x53, x54, x55, x56, x57, x58, x59, x60, x61, x62, x63 | - From: | - /// A 512-bit vector with 64 `i8` lanes. -); -impl_u!([u8; 64]: u8x64, m8x64 | u8, u64 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31, - x32, x33, x34, x35, x36, x37, x38, x39, x40, x41, x42, x43, x44, x45, x46, x47, - x48, x49, x50, x51, x52, x53, x54, x55, x56, x57, x58, x59, x60, x61, x62, x63 | - From: | - /// A 512-bit vector with 64 `u8` lanes. -); -impl_m!([m8; 64]: m8x64 | i8, u64 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31, - x32, x33, x34, x35, x36, x37, x38, x39, x40, x41, x42, x43, x44, x45, x46, x47, - x48, x49, x50, x51, x52, x53, x54, x55, x56, x57, x58, x59, x60, x61, x62, x63 | - From: | - /// A 512-bit vector mask with 64 `m8` lanes. -); - -impl_i!([i16; 32]: i16x32, m16x32 | i16, u32 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | - From: i8x32, u8x32 | - /// A 512-bit vector with 32 `i16` lanes. -); -impl_u!([u16; 32]: u16x32, m16x32 | u16, u32 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | - From: u8x32 | - /// A 512-bit vector with 32 `u16` lanes. -); -impl_m!([m16; 32]: m16x32 | i16, u32 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, - x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | - From: m8x32 | - /// A 512-bit vector mask with 32 `m16` lanes. -); - -impl_i!([i32; 16]: i32x16, m32x16 | i32, u16 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: i8x16, u8x16, i16x16, u16x16 | - /// A 512-bit vector with 16 `i32` lanes. -); -impl_u!([u32; 16]: u32x16, m32x16 | u32, u16 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: u8x16, u16x16 | - /// A 512-bit vector with 16 `u32` lanes. -); -impl_f!([f32; 16]: f32x16, m32x16 | f32 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: i8x16, u8x16, i16x16, u16x16 | - /// A 512-bit vector with 16 `f32` lanes. -); -impl_m!([m32; 16]: m32x16 | i32, u16 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | - From: m8x16, m16x16 | - /// A 512-bit vector mask with 16 `m32` lanes. -); - -impl_i!([i64; 8]: i64x8, m64x8 | i64, u8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: i8x8, u8x8, i16x8, u16x8, i32x8, u32x8 | - /// A 512-bit vector with 8 `i64` lanes. -); -impl_u!([u64; 8]: u64x8, m64x8 | u64, u8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: u8x8, u16x8, u32x8 | - /// A 512-bit vector with 8 `u64` lanes. -); -impl_f!([f64; 8]: f64x8, m64x8 | f64 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: i8x8, u8x8, i16x8, u16x8, i32x8, u32x8, f32x8 | - /// A 512-bit vector with 8 `f64` lanes. -); -impl_m!([m64; 8]: m64x8 | i64, u8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: m8x8, m16x8, m32x8 | - /// A 512-bit vector mask with 8 `m64` lanes. -); - -impl_i!([i128; 4]: i128x4, m128x4 | i128, u8 | test_v512 | x0, x1, x2, x3 | - From: i8x4, u8x4, i16x4, u16x4, i32x4, u32x4, i64x4, u64x4 | - /// A 512-bit vector with 4 `i128` lanes. -); -impl_u!([u128; 4]: u128x4, m128x4 | u128, u8 | test_v512 | x0, x1, x2, x3 | - From: u8x4, u16x4, u32x4, u64x4 | - /// A 512-bit vector with 4 `u128` lanes. -); -impl_m!([m128; 4]: m128x4 | i128, u8 | test_v512 | x0, x1, x2, x3 | - From: m8x4, m16x4, m32x4, m64x4 | - /// A 512-bit vector mask with 4 `m128` lanes. -); diff --git a/third_party/rust/packed_simd/src/v64.rs b/third_party/rust/packed_simd/src/v64.rs deleted file mode 100644 index bf6b9de610..0000000000 --- a/third_party/rust/packed_simd/src/v64.rs +++ /dev/null @@ -1,66 +0,0 @@ -//! 64-bit wide vector types -#[rustfmt::skip] - -use super::*; - -impl_i!([i8; 8]: i8x8, m8x8 | i8, u8 | test_v64 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: | - /// A 64-bit vector with 8 `i8` lanes. -); -impl_u!([u8; 8]: u8x8, m8x8 | u8, u8 | test_v64 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: | - /// A 64-bit vector with 8 `u8` lanes. -); -impl_m!([m8; 8]: m8x8 | i8, u8 | test_v64 | x0, x1, x2, x3, x4, x5, x6, x7 | - From: m16x8, m32x8 | - /// A 64-bit vector mask with 8 `m8` lanes. -); - -impl_i!([i16; 4]: i16x4, m16x4 | i16, u8 | test_v64 | x0, x1, x2, x3 | - From: i8x4, u8x4 | - /// A 64-bit vector with 4 `i16` lanes. -); -impl_u!([u16; 4]: u16x4, m16x4 | u16, u8 | test_v64 | x0, x1, x2, x3 | - From: u8x4 | - /// A 64-bit vector with 4 `u16` lanes. -); -impl_m!([m16; 4]: m16x4 | i16, u8 | test_v64 | x0, x1, x2, x3 | - From: m8x4, m32x4, m64x4 | - /// A 64-bit vector mask with 4 `m16` lanes. -); - -impl_i!([i32; 2]: i32x2, m32x2 | i32, u8 | test_v64 | x0, x1 | - From: i8x2, u8x2, i16x2, u16x2 | - /// A 64-bit vector with 2 `i32` lanes. -); -impl_u!([u32; 2]: u32x2, m32x2 | u32, u8 | test_v64 | x0, x1 | - From: u8x2, u16x2 | - /// A 64-bit vector with 2 `u32` lanes. -); -impl_m!([m32; 2]: m32x2 | i32, u8 | test_v64 | x0, x1 | - From: m8x2, m16x2, m64x2, m128x2 | - /// A 64-bit vector mask with 2 `m32` lanes. -); -impl_f!([f32; 2]: f32x2, m32x2 | f32 | test_v64 | x0, x1 | - From: i8x2, u8x2, i16x2, u16x2 | - /// A 64-bit vector with 2 `f32` lanes. -); - -/* -impl_i!([i64; 1]: i64x1, m64x1 | i64, u8 | test_v64 | x0 | - From: /*i8x1, u8x1, i16x1, u16x1, i32x1, u32x1*/ | // FIXME: primitive to vector conversion - /// A 64-bit vector with 1 `i64` lanes. -); -impl_u!([u64; 1]: u64x1, m64x1 | u64, u8 | test_v64 | x0 | - From: /*u8x1, u16x1, u32x1*/ | // FIXME: primitive to vector conversion - /// A 64-bit vector with 1 `u64` lanes. -); -impl_m!([m64; 1]: m64x1 | i64, u8 | test_v64 | x0 | - From: /*m8x1, m16x1, m32x1, */ m128x1 | // FIXME: unary small vector types - /// A 64-bit vector mask with 1 `m64` lanes. -); -impl_f!([f64; 1]: f64x1, m64x1 | f64 | test_v64 | x0 | - From: /*i8x1, u8x1, i16x1, u16x1, i32x1, u32x1, f32x1*/ | // FIXME: unary small vector types - /// A 64-bit vector with 1 `f64` lanes. -); -*/ diff --git a/third_party/rust/packed_simd/src/vPtr.rs b/third_party/rust/packed_simd/src/vPtr.rs deleted file mode 100644 index e34cb170eb..0000000000 --- a/third_party/rust/packed_simd/src/vPtr.rs +++ /dev/null @@ -1,34 +0,0 @@ -//! Vectors of pointers -#[rustfmt::skip] - -use crate::*; - -impl_const_p!( - [*const T; 2]: cptrx2, msizex2, usizex2, isizex2 | test_v128 | x0, x1 | From: | - /// A vector with 2 `*const T` lanes -); - -impl_mut_p!( - [*mut T; 2]: mptrx2, msizex2, usizex2, isizex2 | test_v128 | x0, x1 | From: | - /// A vector with 2 `*mut T` lanes -); - -impl_const_p!( - [*const T; 4]: cptrx4, msizex4, usizex4, isizex4 | test_v256 | x0, x1, x2, x3 | From: | - /// A vector with 4 `*const T` lanes -); - -impl_mut_p!( - [*mut T; 4]: mptrx4, msizex4, usizex4, isizex4 | test_v256 | x0, x1, x2, x3 | From: | - /// A vector with 4 `*mut T` lanes -); - -impl_const_p!( - [*const T; 8]: cptrx8, msizex8, usizex8, isizex8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | From: | - /// A vector with 8 `*const T` lanes -); - -impl_mut_p!( - [*mut T; 8]: mptrx8, msizex8, usizex8, isizex8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | From: | - /// A vector with 8 `*mut T` lanes -); diff --git a/third_party/rust/packed_simd/src/vSize.rs b/third_party/rust/packed_simd/src/vSize.rs deleted file mode 100644 index b5d8910061..0000000000 --- a/third_party/rust/packed_simd/src/vSize.rs +++ /dev/null @@ -1,53 +0,0 @@ -//! Vectors with pointer-sized elements - -use crate::codegen::pointer_sized_int::{isize_, usize_}; -use crate::*; - -impl_i!([isize; 2]: isizex2, msizex2 | isize_, u8 | test_v128 | - x0, x1| - From: | - /// A vector with 2 `isize` lanes. -); - -impl_u!([usize; 2]: usizex2, msizex2 | usize_, u8 | test_v128 | - x0, x1| - From: | - /// A vector with 2 `usize` lanes. -); -impl_m!([msize; 2]: msizex2 | isize_, u8 | test_v128 | - x0, x1 | - From: | - /// A vector mask with 2 `msize` lanes. -); - -impl_i!([isize; 4]: isizex4, msizex4 | isize_, u8 | test_v256 | - x0, x1, x2, x3 | - From: | - /// A vector with 4 `isize` lanes. -); -impl_u!([usize; 4]: usizex4, msizex4 | usize_, u8 | test_v256 | - x0, x1, x2, x3| - From: | - /// A vector with 4 `usize` lanes. -); -impl_m!([msize; 4]: msizex4 | isize_, u8 | test_v256 | - x0, x1, x2, x3 | - From: | - /// A vector mask with 4 `msize` lanes. -); - -impl_i!([isize; 8]: isizex8, msizex8 | isize_, u8 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7 | - From: | - /// A vector with 8 `isize` lanes. -); -impl_u!([usize; 8]: usizex8, msizex8 | usize_, u8 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7 | - From: | - /// A vector with 8 `usize` lanes. -); -impl_m!([msize; 8]: msizex8 | isize_, u8 | test_v512 | - x0, x1, x2, x3, x4, x5, x6, x7 | - From: | - /// A vector mask with 8 `msize` lanes. -); |